US 9,250,811 B1Grant
Data write caching for sequentially written media
Issue Date:2016-02-02
•26 Claims
•13 Drawing Sheets
Abstract
Techniques for implementing a data queuing and/or caching scheme for optimizing data storage are described herein. Data write requests are received and processed by at least queuing the requests and/or associated data for recording upon one or more data storage devices. The order within the queue, as well as the order in which the queued requests are serviced, may, in some embodiments, be optimized. The stored data are verified by determining the position of a write pointer implemented by the one or more data storage devices relative to the contents and/or position of the queued data requests.
Metadata
Assignee
- Amazon Technologies, Inc.
Inventor
- Kestutis Patiejunas
Application Information
Application Number:US 13/570,005
Filing Date:2012-08-08
Priority Date:2012-08-08
Art Unit:2139
Classifications
IPC:
G06F3/06G11B5/012
Field of Search:
G06F 3/0656G06F 3/0619G06F 3/0674
Patent Drawings (13 sheets)
Description
Cross-Reference to Related Applications
[0001] This application incorporates by reference for all purposes the full disclosure of U.S. patent application Ser. No. 13/569,984, filed concurrently herewith, entitled “LOG-BASED DATA STORAGE ON SEQUENTIALLY WRITTEN MEDIA,” U.S. patent application Ser. No. 13/570,057, filed concurrently herewith, entitled “DATA STORAGE MANAGEMENT FOR SEQUENTIALLY WRITTEN MEDIA,” U.S. patent application Ser. No. 13/569,665, filed concurrently herewith, entitled “DATA STORAGE INVENTORY INDEXING,” U.S. patent application Ser. No. 13/570,030, filed concurrently herewith, entitled “PROGRAMMABLE CHECKSUM CALCULATIONS ON DATA STORAGE DEVICES,” U.S. patent application Ser. No. 13/569,994, filed concurrently herewith, entitled “ARCHIVAL DATA IDENTIFICATION,” U.S. patent application Ser. No. 13/570,029, filed concurrently herewith, entitled “ARCHIVAL DATA ORGANIZATION AND MANAGEMENT,” U.S. patent application Ser. No. 13/570,092, filed concurrently herewith, entitled “ARCHIVAL DATA FLOW MANAGEMENT,” U.S. patent application Ser. No. 13/570,088, filed concurrently herewith, entitled “ARCHIVAL DATA STORAGE SYSTEM,” U.S. patent application Ser. No. 13/569,591, filed concurrently herewith, entitled “DATA STORAGE POWER MANAGEMENT,” U.S. patent application Ser. No. 13/569,714, filed concurrently herewith, entitled “DATA STORAGE SPACE MANAGEMENT,” U.S. patent application Ser. No. 13/570,074, filed concurrently herewith, entitled “DATA STORAGE APPLICATION PROGRAMMING INTERFACE,” and U.S. patent application Ser. No. 13/570,151, filed concurrently herewith, entitled “DATA STORAGE INTEGRITY VALIDATION.”
Background
[0002] In the typical course of electronic communication, increasing amounts of data are generated and stored. As the demands for storage space increase, storage density becomes increasingly important. Shingled magnetic recording (SMR) techniques have been developed as one way to increase storage density at a device level, but devices using SMR techniques have certain operational characteristics that may be taken into account when they are deployed in data storage systems. For example, it may be impractical to write data to SMR devices in a non-sequential fashion, similarly to many tape-based storage devices. On such devices where non-sequential writes may be impractical, data may be written such that one or more portions of data are written contiguously upon the device. When such portions of data are associated with disparate write requests, however, it may be difficult, burdensome and/or inefficient to locate a location on the device to which each successive write should be committed so that they are contiguously positioned.
Brief Description of the Drawings
[0003] Various embodiments in accordance with the present disclosure will be described with reference to the drawings, in which:
[0004] FIG. 1 illustrates an example environment in which a data write verification scheme may be implemented in accordance with some embodiments;
[0005] FIG. 2 illustrates an example environment in which archival data storage services may be implemented, in accordance with at least one embodiment;
[0006] FIG. 3 illustrates an interconnection network in which components of an archival data storage system may be connected, in accordance with at least one embodiment;
[0007] FIG. 4 illustrates an interconnection network in which components of an archival data storage system may be connected, in accordance with at least one embodiment;
[0008] FIG. 5 illustrates an example process for storing data, in accordance with at least one embodiment;
[0009] FIG. 6 illustrates an example process for retrieving data, in accordance with at least one embodiment;
[0010] FIG. 7 illustrates an example process for deleting data, in accordance with at least one embodiment;
[0011] FIG. 8 illustrates a layout of an example data storage medium, also referred to as a data storage media, in which various embodiments can be implemented;
[0012] FIG. 9 illustrates an example data storage format in accordance with at least one embodiment;
[0013] FIG. 10 illustrates an example process for caching, writing and/or verifying data in accordance with at least one embodiment;
[0014] FIG. 11 illustrates an example process for caching, writing and/or verifying data on data storage systems in accordance with at least one embodiment; and
[0015] FIG. 12 illustrates an environment in which various embodiments can be implemented.
Detailed Description
[0016] In the following description, various embodiments will be described. For purposes of explanation, specific configurations and details are set forth in order to provide a thorough understanding of the embodiments. However, it will also be apparent to one skilled in the art that the embodiments may be practiced without the specific details. Furthermore, well-known features may be omitted or simplified in order not to obscure the embodiment being described.
[0017] Techniques described and suggested herein include methods, systems and processes for storing and managing data on append-only data storage media and devices (i.e., those which require writes to be sequential in nature). In particular, techniques for pipelining, caching and/or verifying stored data and data to be stored is described herein. Such techniques may take advantage of, or modify, an implemented scheme for storing and managing data on sequentially written storage devices, where sequential continuity of written data is of importance. For example, it may be impractical to write to certain types of media, such as SMR hard disk media, tape media and the like, in a non-sequential fashion, due to physical and/or logical constraints of their implementation. As another example, while certain types of media, e.g., solid state storage devices and electromechanical hard disk drives using conventional media, may not have such constraints, a system, device or implementation using such media may be configured such that sequential writes would be desirable (for example, to improve data throughput or decrease data fragmentation). A log-based storage design, in some embodiments, allows for improved data write tracking and indexing and increase consistency of data stored on such media, implementations or devices, e.g., as used in a larger data storage system consisting of a plurality of devices containing such media, such as a data storage system used for archival purposes.
[0018] In some embodiments, each sequential portion of data written to devices or media applying the log-based storage design is both preceded and succeeded by records containing information relating to the contents of the written data. Such records are linked together and, in some embodiments, may be traversed iteratively by an device processing the records so as to develop a representation of a state of the entire device or media. In some embodiments, some of the records may be located at known locations upon the device or media such that they may be directly located by an external entity seeking a certain subset of the data stored thereon. Some of these directly locatable records include, in some embodiments, indexing information for some or all of the data stored on the device or media sequentially preceding that record. Some combination of the approaches described may be used by a deploying system.
[0019] In some embodiments, media or devices deploying the log-based storage design may be capable of performing random reads, i.e., able to seek to and read specific locations without first traversing to the location sequentially. Data stored in such a log-based storage scheme is, as previously described, indexed in various ways in some or all of the associated records, and thus the indexes may be read, processed or cached in order to more quickly locate a specific portion of data for which the index has an entry. It is contemplated that, in some embodiments, some or all of the records may contain metadata to further improve data search efficiency or to decrease the amount of overhead necessary to, e.g., cache an index of such data.
[0020] For write append-only media and devices using such media, deletion of specific data previously stored thereon may be impractical or even impossible by conventional means, e.g., simply by locating the data to be deleted and directly overwriting it, or removing an index reference to such data and marking the associated data location to be available for further writes. In embodiments where the log-based storage design described herein is applied to write append-only media, deletion requests may be serviced by adding information to the next record sequentially written to the media, including, for example, a reference to the location of the originally stored data upon the media and information indicating that the data should be omitted. When the media or device is partially or fully traversed, i.e., to acquire a state of the data stored thereon, such data will, in some embodiments, be noted by the acquiring entity as unavailable, and thus, effectively deleted.
[0021] On sequentially written devices that, for example, use a log-based storage design or other indexing methods to store, manage and/or read data stored thereon, an write location identifier (also referred to as a write pointer) that indicates a current or sequentially furthest write position of the device may be implemented. Such a write pointer may be implemented by, for example, the data storage device upon which the data are stored, and may, in some embodiments, be exposed upon request of an connected entity external to the data storage device. When implemented by the data storage device, the write pointer may be deployed in such a way that minimizes interruption to write and/or read operations to and/or from the data storage device. A write queue or caching scheme may be used in conjunction with a write pointer implementation such that, in embodiments where the data storage device verifies that given data have been successfully committed to the associated media prior to updating the location of the write pointer, the location of the write pointer as queried may be used by the queuing or caching entity to determine whether or how much of the data residing in the queue or cache has successfully been committed to the data storage device.
[0022] FIG. 1 illustrates an example environment 100 in which a data queue and/or cache interacts with an implemented write pointer in accordance with some embodiments. The particular illustration shown in FIG. 1 is simplified for the purpose of illustration and components of the scheme may vary in order, content and/or quantity. As illustrated, a data storage node 102 implements or is connected to a data write queue 104 containing or referring to data, associated with one or more write requests, that, at least in part, are to be committed (i.e., causing the data to be written) to a given data storage device 106. The data storage node may be among a plurality of such nodes, which may connect to a plurality of such data storage devices, such as described in connection with the archival data storage system 206 of FIGS. 2-7 . The data write queue may be, for example, the job request queue 242 described in connection with at least FIG. 2 below, or, in some embodiments, may be a lower level cache or queue associated with an entity similar to the job request queue. Data write requests may be queued, for example, in response to receiving multiple data objects to be uploaded to a data storage system including the data storage node, data write queue and data storage device. In some embodiments, the associated data may be committed to the data storage device, for example, when the data storage node and/or the data storage device indicate that they are prepared to accept additional data for storage, and/or in response to a determination by a connected entity (for example, the storage allocator 256 described in connection with FIG. 2 ) that one or more data locations upon the data storage device have been selected as the target location for the queued data. Data and/or associated data write requests may be received from any appropriate connected entity, such as the various components of the archival data storage system 206. The data storage node may order the requests within the data write queue in any appropriate order, such as first-in-first-out, first-in-last-out or some other order (for example, by determining that one or more data write requests pertain to a given object or portion of data, and subsequently optimizing the order of the associated data writes to minimize fragmentation and/or maximize performance).
[0023] As the data are sent to the data storage device, the data storage device sends information relating to a write pointer 108, as may be implemented by a controller, firmware or similar entity 110 at least connected to the media of the data storage device. In some embodiments, the write pointer (and associated information) may be stored upon the data storage device media, within volatile and/or non-volatile memory associated with the controller and/or firmware of the data storage device, or both. In embodiments where the write pointer and associated information is stored upon the data storage media, it is contemplated that the location may be cached, held and/or updated at some interval within a volatile and/or non-volatile memory of (or associated with) the controller and/or firmware. Such implementations may be advantageous, for example, to allow for low latency and/or highly efficient retrieval of the write pointer (e.g., from the volatile and/or non-volatile memory rather than, or in addition to, from the media) during operation of the data storage device. As may be contemplated, in embodiments and/or situations where the storage of such a write pointer is discontinued (e.g., after power down of the data storage device if the write pointer is being held in a volatile memory), upon restart, the write pointer may be retrieved from the media and restored to the volatile and/or non-volatile memory. The write pointer information may be sent in response to an external query, e.g., from a data storage node, or may be sent automatically by the implementing data storage device according to a schedule (e.g., every N seconds) or when certain criteria are met, for example, upon every N writes committed to the drive. The write pointer may be implemented in such a way, for example by persisting information relating to the write position with which the write pointer is aligned within low-latency storage such as memory, as may be accessible using a controller of the data storage device, that requests for the write pointer's location do not hamper, or at least only minimally affect, other operations of the data storage device and/or the data storage node. In some embodiments, the data storage device may continuously update the write pointer information as writes occur, and persist the write pointer information within a controller or adapter for the drive for external requests and or consumption. In some embodiments, the write pointer and related information are externally accessible via, for example, a programmatic interface such as an application programming interface (API).
[0024] In embodiments where the data storage device is written sequentially, as would generally be the case in embodiments where the data storage device uses SMR techniques (described in further detail below in connection with at least FIG. 10 ) or other techniques having similar write characteristics, such as techniques used in connection with tape drives, the data storage node may use the location of the write pointer to determine what portion of the data within the data write queue has successfully been written. In the illustrated example, data elements 1 through 7 have been queued, of which data elements 1 through 5 have been successfully committed to the data storage device. The write pointer information provided to the data storage node reflects the location of the write pointer at the data location associated with data element 5, and thus the data storage node may deduce that data elements 6 and 7 still remain in the queue and have not yet been written. Such a scheme may have performance and efficiency benefits over preemptive flushing of the queue or cache to, for example, ensure that all data within the queue or cache have been committed, as such flushing or commits may delay or slow down pending write and/or read requests being processed by the data storage device, as well as cause the data to be written in a suboptimal sequence. The performance and efficiency benefits thus described may be further enhanced in embodiments where the write pointer and related information may be queried and/or provided with little or no effect upon current read and/or write operations of the data storage device. Additionally, in some embodiments, as the write pointer reflects that data has successfully been written to the device, less data verification may be required relative to, for example, techniques where a cache or queue is “pushed” or similarly flushed to the device.
[0025] FIG. 2 illustrates an example environment 200 in which an archival data storage system may be implemented, in accordance with at least one embodiment. One or more customers 202 connect, via a network 204, to an archival data storage system 206. As implied above, unless otherwise clear from context, the term “customer” refers to the system(s) of a customer entity (such as an individual, company or other organization) that utilizes data storage services described herein. Such systems may include datacenters, mainframes, individual computing devices, distributed computing environments and customer-accessible instances thereof or any other system capable of communicating with the archival data storage system. In some embodiments, a customer may refer to a machine instance (e.g., with direct hardware access) or virtual instance of a distributed computing system provided by a computing resource provider that also provides the archival data storage system. In some embodiments, the archival data storage system is integral to the distributed computing system and may include or be implemented by an instance, virtual or machine, of the distributed computing system. In various embodiments, network 204 may include the Internet, a local area network (“LAN”), a wide area network (“WAN”), a cellular data network and/or other data network.
[0026] In an embodiment, archival data storage system 206 provides a multi-tenant or multi-customer environment where each tenant or customer may store, retrieve, delete or otherwise manage data in a data storage space allocated to the customer. In some embodiments, an archival data storage system 206 comprises multiple subsystems or “planes” that each provides a particular set of services or functionalities. For example, as illustrated in FIG. 2 , archival data storage system 206 includes front end 208, control plane for direct I/O 210, common control plane 212, data plane 214 and metadata plane 216. Each subsystem or plane may comprise one or more components that collectively provide the particular set of functionalities. Each component may be implemented by one or more physical and/or logical computing devices, such as computers, data storage devices and the like. Components within each subsystem may communicate with components within the same subsystem, components in other subsystems or external entities such as customers. At least some of such interactions are indicated by arrows in FIG. 2 . In particular, the main bulk data transfer paths in and out of archival data storage system 206 are denoted by bold arrows. It will be appreciated by those of ordinary skill in the art that various embodiments may have fewer or a greater number of systems, subsystems and/or subcomponents than are illustrated in FIG. 2 . Thus, the depiction of environment 200 in FIG. 2 should be taken as being illustrative in nature and not limiting to the scope of the disclosure.
[0027] In the illustrative embodiment, front end 208 implements a group of services that provides an interface between the archival data storage system 206 and external entities, such as one or more customers 202 described herein. In various embodiments, front end 208 provides an application programming interface (“API”) to enable a user to programmatically interface with the various features, components and capabilities of the archival data storage system. Such APIs may be part of a user interface that may include graphical user interfaces (GUIs). Web-based interfaces, programmatic interfaces such as application programming interfaces (APIs) and/or sets of remote procedure calls (RPCs) corresponding to interface elements, messaging interfaces in which the interface elements correspond to messages of a communication protocol, and/or suitable combinations thereof.
[0028] Capabilities provided by archival data storage system 206 may include data storage, data retrieval, data deletion, metadata operations, configuration of various operational parameters and the like. Metadata operations may include requests to retrieve catalogs of data stored for a particular customer, data recovery requests, job inquires and the like. Configuration APIs may allow customers to configure account information, audit logs, policies, notifications settings and the like. A customer may request the performance of any of the above operations by sending API requests to the archival data storage system. Similarly, the archival data storage system may provide responses to customer requests. Such requests and responses may be submitted over any suitable communications protocol, such as Hypertext Transfer Protocol (“HTTP”), File Transfer Protocol (“FTP”) and the like, in any suitable format, such as REpresentational State Transfer (“REST”), Simple Object Access Protocol (“SOAP”) and the like. The requests and responses may be encoded, for example, using Base64 encoding, encrypted with a cryptographic key or the like.
[0029] In some embodiments, archival data storage system 206 allows customers to create one or more logical structures such as a logical data containers in which to store one or more archival data objects. As used herein, data object is used broadly and does not necessarily imply any particular structure or relationship to other data. A data object may be, for instance, simply a sequence of bits. Typically, such logical data structures may be created to meeting certain business requirements of the customers and are independently of the physical organization of data stored in the archival data storage system. As used herein, the term “logical data container” refers to a grouping of data objects. For example, data objects created for a specific purpose or during a specific period of time may be stored in the same logical data container. Each logical data container may include nested data containers or data objects and may be associated with a set of policies such as size limit of the container, maximum number of data objects that may be stored in the container, expiration date, access control list and the like. In various embodiments, logical data containers may be created, deleted or otherwise modified by customers via API requests, by a system administrator or by the data storage system, for example, based on configurable information. For example, the following HTTP PUT request may be used, in an embodiment, to create a logical data container with name “logical-container-name” associated with a customer identified by an account identifier “accountId”.
-
- [0030] PUT /{accountId}/logical-container-name HTTP/1.1
[0031] In an embodiment, archival data storage system 206 provides the APIs for customers to store data objects into logical data containers. For example, the following HTTP POST request may be used, in an illustrative embodiment, to store a data object into a given logical container. In an embodiment, the request may specify the logical path of the storage location, data length, reference to the data payload, a digital digest of the data payload and other information. In one embodiment, the APIs may allow a customer to upload multiple data objects to one or more logical data containers in one request. In another embodiment where the data object is large, the APIs may allow a customer to upload the data object in multiple parts, each with a portion of the data object.
-
- [0032] POST /{accountId}/logical-container-name/data HTTP/1.1
- [0033] Content-Length: 1128192
- [0034] x-ABC-data-description: “annual-result-2012.xls”
- [0035] x-ABC-md5-tree-hash: 634d9a0688aff95c
[0036] In response to a data storage request, in an embodiment, archival data storage system 206 provides a data object identifier if the data object is stored successfully. Such data object identifier may be used to retrieve, delete or otherwise refer to the stored data object in subsequent requests. In some embodiments, such as data object identifier may be “self-describing” in that it includes (for example, with or without encryption) storage location information that may be used by the archival data storage system to locate the data object without the need for a additional data structures such as a global namespace key map. In addition, in some embodiments, data object identifiers may also encode other information such as payload digest, error-detection code, access control data and the other information that may be used to validate subsequent requests and data integrity. In some embodiments, the archival data storage system stores incoming data in a transient durable data store before moving it archival data storage. Thus, although customers may perceive that data is persisted durably at the moment when an upload request is completed, actual storage to a long-term persisted data store may not commence until sometime later (e.g., 12 hours later). In some embodiments, the timing of the actual storage may depend on the size of the data object, the system load during a diurnal cycle, configurable information such as a service-level agreement between a customer and a storage service provider and other factors.
[0037] In some embodiments, archival data storage system 206 provides the APIs for customers to retrieve data stored in the archival data storage system. In such embodiments, a customer may initiate a job to perform the data retrieval and may learn the completion of the job by a notification or by polling the system for the status of the job. As used herein, a “job” refers to a data-related activity corresponding to a customer request that may be performed temporally independently from the time the request is received. For example, a job may include retrieving, storing and deleting data, retrieving metadata and the like. A job may be identified by a job identifier that may be unique, for example, among all the jobs for a particular customer. For example, the following HTTP POST request may be used, in an illustrative embodiment, to initiate a job to retrieve a data object identified by a data object identifier “dataObjectId.” In other embodiments, a data retrieval request may request the retrieval of multiple data objects, data objects associated with a logical data container and the like.
-
- [0038] POST /{accountId}/logical-data-container-name/data/{dataObjectId} HTTP/1.1
[0039] In response to the request, in an embodiment, archival data storage system 206 provides a job identifier job-id,” that is assigned to the job in the following response. The response provides, in this example, a path to the storage location where the retrieved data will be stored.
-
- [0040] HTTP/1.1 202 ACCEPTED
- [0041] Location: /{accountId}/logical-data-container-name/jobs/{job-id}
[0042] At any given point in time, the archival data storage system may have many jobs pending for various data operations. In some embodiments, the archival data storage system may employ job planning and optimization techniques such as batch processing, load balancing, job coalescence and the like, to optimize system metrics such as cost, performance, scalability and the like. In some embodiments, the timing of the actual data retrieval depends on factors such as the size of the retrieved data, the system load and capacity, active status of storage devices and the like. For example, in some embodiments, at least some data storage devices in an archival data storage system may be activated or inactivated according to a power management schedule, for example, to reduce operational costs. Thus, retrieval of data stored in a currently active storage device (such as a rotating hard drive) may be faster than retrieval of data stored in a currently inactive storage device (such as a spinned-down hard drive).
[0043] In an embodiment, when a data retrieval job is completed, the retrieved data is stored in a staging data store and made available for customer download. In some embodiments, a customer is notified of the change in status of a job by a configurable notification service. In other embodiments, a customer may learn of the status of a job by polling the system using a job identifier. The following HTTP GET request may be used, in an embodiment, to download data that is retrieved by a job identified by “job-id,” using a download path that has been previously provided.
-
- [0044] GET /{accountId}/logical-data-container-name/jobs/{job-id}/output HTTP/1.1
[0045] In response to the GET request, in an illustrative embodiment, archival data storage system 206 may provide the retrieved data in the following HTTP response, with a tree-hash of the data for verification purposes.
-
- [0046] HTTP/1.1 200 OK
- [0047] Content-Length: 1128192
- [0048] x-ABC-archive-description: “retrieved stuff”
- [0049] x-ABC-md5-tree-hash: 693d9a7838aff95c
- [0050] [1112192 bytes of user data follows]
[0051] In an embodiment, a customer may request the deletion of a data object stored in an archival data storage system by specifying a data object identifier associated with the data object. For example, in an illustrative embodiment, a data object with data object identifier “dataObjectId” may be deleted using the following HTTP request. In another embodiment, a customer may request the deletion of multiple data objects such as those associated with a particular logical data container.
-
- [0052] DELETE /{accountId}/logical-data-container-name/data/{dataObjectId}HTTP/1.1
[0053] In various embodiments, data objects may be deleted in response to a customer request or may be deleted automatically according to a user-specified or default expiration date. In some embodiments, data objects may be rendered inaccessible to customers upon an expiration time but remain recoverable during a grace period beyond the expiration time. In various embodiments, the grace period may be based on configurable information such as customer configuration, service-level agreement terms and the like. In some embodiments, a customer may be provided the abilities to query or receive notifications for pending data deletions and/or cancel one or more of the pending data deletions. For example, in one embodiment, a customer may set up notification configurations associated with a logical data container such that the customer will receive notifications of certain events pertinent to the logical data container. Such events may include the completion of a data retrieval job request, the completion of metadata request, deletion of data objects or logical data containers and the like.
[0054] In an embodiment, archival data storage system 206 also provides metadata APIs for retrieving and managing metadata such as metadata associated with logical data containers. In various embodiments, such requests may be handled asynchronously (where results are returned later) or synchronously (where results are returned immediately).
[0055] Still referring to FIG. 2 , in an embodiment, at least some of the API requests discussed above are handled by API request handler 218 as part of front end 208. For example, API request handler 218 may decode and/or parse an incoming API request to extract information, such as uniform resource identifier (“URI”), requested action and associated parameters, identity information, data object identifiers and the like. In addition, API request handler 218 invoke other services (described below), where necessary, to further process the API request.
[0056] In an embodiment, front end 208 includes an authentication service 220 that may be invoked, for example, by API handler 218, to authenticate an API request. For example, in some embodiments, authentication service 220 may verify identity information submitted with the API request such as username and password Internet Protocol (“IP) address, cookies, digital certificate, digital signature and the like. In other embodiments, authentication service 220 may require the customer to provide additional information or perform additional steps to authenticate the request, such as required in a multifactor authentication scheme, under a challenge-response authentication protocol and the like.
[0057] In an embodiment, front end 208 includes an authorization service 222 that may be invoked, for example, by API handler 218, to determine whether a requested access is permitted according to one or more policies determined to be relevant to the request. For example, in one embodiment, authorization service 222 verifies that a requested access is directed to data objects contained in the requestor's own logical data containers or which the requester is otherwise authorized to access. In some embodiments, authorization service 222 or other services of front end 208 may check the validity and integrity of a data request based at least in part on information encoded in the request, such as validation information encoded by a data object identifier.
[0058] In an embodiment, front end 208 includes a metering service 224 that monitors service usage information for each customer such as data storage space used, number of data objects stored, data requests processed and the like. In an embodiment, front end 208 also includes accounting service 226 that performs accounting and billing-related functionalities based, for example, on the metering information collected by the metering service 224, customer account information and the like. For example, a customer may be charged a fee based on the storage space used by the customer, size and number of the data objects, types and number of requests submitted, customer account type, service level agreement the like.
[0059] In an embodiment, front end 208 batch processes some or all incoming requests. For example, front end 208 may wait until a certain number of requests has been received before processing (e.g., authentication, authorization, accounting and the like) the requests. Such a batch processing of incoming requests may be used to gain efficiency.
[0060] In some embodiments, front end 208 may invoke services provided by other subsystems of the archival data storage system to further process an API request. For example, front end 208 may invoke services in metadata plane 216 to fulfill metadata requests. For another example, front end 208 may stream data in and out of control plane for direct I/O 210 for data storage and retrieval requests, respectively.
[0061] Referring now to control plane for direct I/O 210 illustrated in FIG. 2 , in various embodiments, control plane for direct I/O 210 provides services that create, track and manage jobs created as a result of customer requests. As discussed above, a job refers to a customer-initiated activity that may be performed asynchronously to the initiating request, such as data retrieval, storage, metadata queries or the like. In an embodiment, control plane for direct I/O 210 includes a job tracker 230 that is configured to create job records or entries corresponding to customer requests, such as those received from API request handler 218, and monitor the execution of the jobs. In various embodiments, a job record may include information related to the execution of a job such as a customer account identifier, job identifier, data object identifier, reference to payload data cache 228 (described below), job status, data validation information and the like. In some embodiments, job tracker 230 may collect information necessary to construct a job record from multiple requests. For example, when a large amount of data is requested to be stored, data upload may be broken into multiple requests, each uploading a portion of the data. In such a case, job tracker 230 may maintain information to keep track of the upload status to ensure that all data parts have been received before a job record is created. In some embodiments, job tracker 230 also obtains a data object identifier associated with the data to be stored and provides the data object identifier, for example, to a front end service to be returned to a customer. In an embodiment, such data object identifier may be obtained from data plane 214 services such as storage node manager 244, storage node registrar 248, and the like, described below.
[0062] In some embodiments, control plane for direct I/O 210 includes a job tracker store 232 for storing job entries or records. In various embodiments, job tracker store 232 may be implemented by a NoSQL data management system, such as a key-value data store, a relational database management system (“RDBMS”) or any other data storage system. In some embodiments, data stored in job tracker store 232 may be partitioned to enable fast enumeration of jobs that belong to a specific customer, facilitate efficient bulk record deletion, parallel processing by separate instances of a service and the like. For example, job tracker store 232 may implement tables that are partitioned according to customer account identifiers and that use job identifiers as range keys. In an embodiment, job tracker store 232 is further sub-partitioned based on time (such as job expiration time) to facilitate job expiration and cleanup operations. In an embodiment, transactions against job tracker store 232 may be aggregated to reduce the total number of transactions. For example, in some embodiments, a job tracker 230 may perform aggregate multiple jobs corresponding to multiple requests into one single aggregated job before inserting it into job tracker store 232.
[0063] In an embodiment, job tracker 230 is configured to submit the job for further job scheduling and planning, for example, by services in common control plane 212. Additionally, job tracker 230 may be configured to monitor the execution of jobs and update corresponding job records in job tracker store 232 as jobs are completed. In some embodiments, job tracker 230 may be further configured to handle customer queries such as job status queries. In some embodiments, job tracker 230 also provides notifications of job status changes to customers or other services of the archival data storage system. For example, when a data retrieval job is completed, job tracker 230 may cause a customer to be notified (for example, using a notification service) that data is available for download. As another example, when a data storage job is completed, job tracker 230 may notify a cleanup agent 234 to remove payload data associated with the data storage job from a transient payload data cache 228, described below.
[0064] In an embodiment, control plane for direct I/O 210 includes a payload data cache 228 for providing transient data storage services for payload data transiting between data plane 214 and front end 208. Such data includes incoming data pending storage and outgoing data pending customer download. As used herein, transient data store is used interchangeably with temporary or staging data store to refer to a data store that is used to store data objects before they are stored in an archival data storage described herein or to store data objects that are retrieved from the archival data storage. A transient data store may provide volatile or non-volatile (durable) storage. In most embodiments, while potentially usable for persistently storing data, a transient data store is intended to store data for a shorter period of time than an archival data storage system and may be less cost-effective than the data archival storage system described herein. In one embodiment, transient data storage services provided for incoming and outgoing data may be differentiated. For example, data storage for the incoming data, which is not yet persisted in archival data storage, may provide higher reliability and durability than data storage for outgoing (retrieved) data, which is already persisted in archival data storage. In another embodiment, transient storage may be optional for incoming data, that is, incoming data may be stored directly in archival data storage without being stored in transient data storage such as payload data cache 228, for example, when there is the system has sufficient bandwidth and/or capacity to do so.
[0065] In an embodiment, control plane for direct I/O 210 also includes a cleanup agent 234 that monitors job tracker store 232 and/or payload data cache 228 and removes data that is no longer needed. For example, payload data associated with a data storage request may be safely removed from payload data cache 228 after the data is persisted in permanent storage (e.g., data plane 214). On the reverse path, data staged for customer download may be removed from payload data cache 228 after a configurable period of time (e.g., 30 days since the data is staged) or after a customer indicates that the staged data is no longer needed.
[0066] In some embodiments, cleanup agent 234 removes a job record from job tracker store 232 when the job status indicates that the job is complete or aborted. As discussed above, in some embodiments, job tracker store 232 may be partitioned to enable to enable faster cleanup. In one embodiment where data is partitioned by customer account identifiers, cleanup agent 234 may remove an entire table that stores jobs for a particular customer account when the jobs are completed instead of deleting individual jobs one at a time. In another embodiment where data is further sub-partitioned based on job expiration time cleanup agent 234 may bulk-delete a whole partition or table of jobs after all the jobs in the partition expire. In other embodiments, cleanup agent 234 may receive instructions or control messages (such as indication that jobs are completed) from other services such as job tracker 230 that cause the cleanup agent 234 to remove job records from job tracker store 232 and/or payload data cache 228.
[0067] Referring now to common control plane 212 illustrated in FIG. 2 . In various embodiments, common control plane 212 provides a queue-based load leveling service to dampen peak to average load levels (jobs) coming from control plane for I/O 210 and to deliver manageable workload to data plane 214. In an embodiment, common control plane 212 includes a job request queue 236 for receiving jobs created by job tracker 230 in control plane for direct I/O 210, described above, a storage node manager job store 240 from which services from data plane 214 (e.g., storage node managers 244) pick up work to execute and a request balancer 238 for transferring job items from job request queue 236 to storage node manager job store 240 in an intelligent manner.
[0068] In an embodiment, job request queue 236 provides a service for inserting items into and removing items from a queue (e.g., first-in-first-out (FIFO) or first-in-last-out (FILO)), a set or any other suitable data structure. Job entries in the job request queue 236 may be similar to or different from job records stored in job tracker store 232, described above.
[0069] In an embodiment, common control plane 212 also provides a durable high efficiency job store, storage node manager job store 240, that allows services from data plane 214 (e.g., storage node manager 244, anti-entropy watcher 252) to perform job planning optimization, check pointing and recovery. For example, in an embodiment, storage node manager job store 240 allows the job optimization such as batch processing, operation coalescing and the like by supporting scanning, querying, sorting or otherwise manipulating and managing job items stored in storage node manager job store 240. In an embodiment, a storage node manager 244 scans incoming jobs and sort the jobs by the type of data operation (e.g., read, write or delete), storage locations (e.g., volume, disk), customer account identifier and the like. The storage node manager 244 may then reorder, coalesce, group in batches or otherwise manipulate and schedule the jobs for processing. For example, in one embodiment, the storage node manager 244 may batch process all the write operations before all the read and delete operations. In another embodiment, the storage node manager 244 may perform operation coalescing. For another example, the storage node manager 244 may coalesce multiple retrieval jobs for the same object into one job or cancel a storage job and a deletion job for the same data object where the deletion job comes after the storage job.
[0070] In an embodiment, storage node manager job store 240 is partitioned, for example, based on job identifiers, so as to allow independent processing of multiple storage node managers 244 and to provide even distribution of the incoming workload to all participating storage node managers 244. In various embodiments, storage node manager job store 240 may be implemented by a NoSQL data management system, such as a key-value data store, a RDBMS or any other data storage system.
[0071] In an embodiment, request balancer 238 provides a service for transferring job items from job request queue 236 to storage node manager job store 240 so as to smooth out variation in workload and to increase system availability. For example, request balancer 238 may transfer job items from job request queue 236 at a lower rate or at a smaller granularity when there is a surge in job requests coming into the job request queue 236 and vice versa when there is a lull in incoming job requests so as to maintain a relatively sustainable level of workload in the storage node manager store 240. In some embodiments, such sustainable level of workload is around the same or below the average workload of the system.
[0072] In an embodiment, job items that are completed are removed from storage node manager job store 240 and added to the job result queue 242. In an embodiment, data plane 214 services (e.g., storage node manager 244) are responsible for removing the job items from the storage node manager job store 240 and adding them to job result queue 242. In some embodiments, job request queue 242 is implemented in a similar manner as job request queue 235, discussed above.
[0073] Referring now to data plane 214 illustrated in FIG. 2 . In various embodiments, data plane 214 provides services related to long-term archival data storage, retrieval and deletion, data management and placement, anti-entropy operations and the like. In various embodiments, data plane 214 may include any number and type of storage entities such as data storage devices (such as tape drives, hard disk drives, solid state devices, and the like), storage nodes or servers, datacenters and the like. Such storage entities may be physical, virtual or any abstraction thereof (e.g., instances of distributed storage and/or computing systems) and may be organized into any topology, including hierarchical or tiered topologies. Similarly, the components of the data plane may be dispersed, local or any combination thereof. For example, various computing or storage components may be local or remote to any number of datacenters, servers or data storage devices, which in turn may be local or remote relative to one another. In various embodiments, physical storage entities may be designed for minimizing power and cooling costs by controlling the portions of physical hardware that are active (e.g., the number of hard drives that are actively rotating). In an embodiment, physical storage entities implement techniques, such as Shingled Magnetic Recording (SMR), to increase storage capacity.
[0074] In an environment illustrated by FIG. 2 , one or more storage node managers 244 each controls one or more storage nodes 246 by sending and receiving data and control messages. Each storage node 246 in turn controls a (potentially large) collection of data storage devices such as hard disk drives. In various embodiments, a storage node manager 244 may communicate with one or more storage nodes 246 and a storage node 246 may communicate with one or more storage node managers 244. In an embodiment, storage node managers 244 are implemented by one or more computing devices that are capable of performing relatively complex computations such as digest computation, data encoding and decoding, job planning and optimization and the like. In some embodiments, storage nodes 246 are implemented by one or more computing devices with less powerful computation capabilities than storage node managers 244. Further, in some embodiments the storage node manager 244 may not be included in the data path. For example, data may be transmitted from the payload data cache 228 directly to the storage nodes 246 or from one or more storage nodes 246 to the payload data cache 228. In this way, the storage node manager 244 may transmit instructions to the payload data cache 228 and/or the storage nodes 246 without receiving the payloads directly from the payload data cache 228 and/or storage nodes 246. In various embodiments, a storage node manager 244 may send instructions or control messages to any other components of the archival data storage system 206 described herein to direct the flow of data.
[0075] In an embodiment, a storage node manager 244 serves as an entry point for jobs coming into and out of data plane 214 by picking job items from common control plane 212 (e.g., storage node manager job store 240), retrieving staged data from payload data cache 228 and performing necessary data encoding for data storage jobs and requesting appropriate storage nodes 246 to store, retrieve or delete data. Once the storage nodes 246 finish performing the requested data operations, the storage node manager 244 may perform additional processing, such as data decoding and storing retrieved data in payload data cache 228 for data retrieval jobs, and update job records in common control plane 212 (e.g., removing finished jobs from storage node manager job store 240 and adding them to job result queue 242).
[0076] In an embodiment, storage node manager 244 performs data encoding according to one or more data encoding schemes before data storage to provide data redundancy, security and the like. Such data encoding schemes may include encryption schemes, redundancy encoding schemes such as erasure encoding, redundant array of independent disks (RAID) encoding schemes, replication and the like. Likewise, in an embodiment, storage node managers 244 performs corresponding data decoding schemes, such as decryption, erasure-decoding and the like, after data retrieval to restore the original data.
[0077] As discussed above in connection with storage node manager job store 240, storage node managers 244 may implement job planning and optimizations such as batch processing, operation coalescing and the like to increase efficiency. In some embodiments, jobs are partitioned among storage node managers so that there is little or no overlap between the partitions. Such embodiments facilitate parallel processing by multiple storage node managers, for example, by reducing the probability of racing or locking.
[0078] In various embodiments, data plane 214 is implemented to facilitate data integrity. For example, storage entities handling bulk data flows such as storage nodes managers 244 and/or storage nodes 246 may validate the digest of data stored or retrieved, check the error-detection code to ensure integrity of metadata and the like.
[0079] In various embodiments, data plane 214 is implemented to facilitate scalability and reliability of the archival data storage system. For example, in one embodiment, storage node managers 244 maintain no or little internal state so that they can be added, removed or replaced with little adverse impact. In one embodiment, each storage device is a self-contained and self-describing storage unit capable of providing information about data stored thereon. Such information may be used to facilitate data recovery in case of data loss. Furthermore, in one embodiment, each storage node 246 is capable of collecting and reporting information about the storage node including the network location of the storage node and storage information of connected storage devices to one or more storage node registrars 248 and/or storage node registrar stores 250. In some embodiments, storage nodes 246 perform such self-reporting at system start up time and periodically provide updated information. In various embodiments, such a self-reporting approach provides dynamic and up-to-date directory information without the need to maintain a global namespace key map or index which can grow substantially as large amounts of data objects are stored in the archival data system.
[0080] In an embodiment, data plane 214 may also include one or more storage node registrars 248 that provide directory information for storage entities and data stored thereon, data placement services and the like. Storage node registrars 248 may communicate with and act as a front end service to one or more storage node registrar stores 250, which provide storage for the storage node registrars 248. In various embodiments, storage node registrar store 250 may be implemented by a NoSQL data management system, such as a key-value data store, a RDBMS or any other data storage system. In some embodiments, storage node registrar stores 250 may be partitioned to enable parallel processing by multiple instances of services. As discussed above, in an embodiment, information stored at storage node registrar store 250 is based at least partially on information reported by storage nodes 246 themselves.
[0081] In some embodiments, storage node registrars 248 provide directory service, for example, to storage node managers 244 that want to determine which storage nodes 246 to contact for data storage, retrieval and deletion operations. For example, given a volume identifier provided by a storage node manager 244, storage node registrars 248 may provide, based on a mapping maintained in a storage node registrar store 250, a list of storage nodes that host volume components corresponding to the volume identifier. Specifically, in one embodiment, storage node registrar store 250 stores a mapping between a list of identifiers of volumes or volume components and endpoints, such as Domain Name System (DNS) names, of storage nodes that host the volumes or volume components.
[0082] As used herein, a “volume” refers to a logical storage space within a data storage system in which data objects may be stored. A volume may be identified by a volume identifier. A volume may reside in one physical storage device (e.g., a hard disk) or span across multiple storage devices. In the latter case, a volume comprises a plurality of volume components each residing on a different storage device. As used herein, a “volume component” refers a portion of a volume that is physically stored in a storage entity such as a storage device. Volume components for the same volume may be stored on different storage entities. In one embodiment, when data is encoded by a redundancy encoding scheme (e.g., erasure coding scheme, RAID, replication), each encoded data component or “shard” may be stored in a different volume component to provide fault tolerance and isolation. In some embodiments, a volume component is identified by a volume component identifier that includes a volume identifier and a shard slot identifier. As used herein, a shard slot identifies a particular shard, row or stripe of data in a redundancy encoding scheme. For example, in one embodiment, a shard slot corresponds to an erasure coding matrix row. In some embodiments, storage node registrar store 250 also stores information about volumes or volume components such as total, used and free space, number of data objects stored and the like.
[0083] In some embodiments, data plane 214 also includes a storage allocator 256 for allocating storage space (e.g., volumes) on storage nodes to store new data objects, based at least in part on information maintained by storage node registrar store 250, to satisfy data isolation and fault tolerance constraints. In some embodiments, storage allocator 256 requires manual intervention.
[0084] In some embodiments, data plane 214 also includes an anti-entropy watcher 252 for detecting entropic effects and initiating anti-entropy correction routines. For example, anti-entropy watcher 252 may be responsible for monitoring activities and status of all storage entities such as storage nodes, reconciling live or actual data with maintained data and the like. In various embodiments, entropic effects include, but are not limited to, performance degradation due to data fragmentation resulting from repeated write and rewrite cycles, hardware wear (e.g., of magnetic media), data unavailability and/or data loss due to hardware/software malfunction, environmental factors, physical destruction of hardware, random chance or other causes. Anti-entropy watcher 252 may detect such effects and in some embodiments may preemptively and/or reactively institute anti-entropy correction routines and/or policies.
[0085] In an embodiment, anti-entropy watcher 252 causes storage nodes 246 to perform periodic anti-entropy scans on storage devices connected to the storage nodes. Anti-entropy watcher 252 may also inject requests in job request queue 236 (and subsequently job result queue 242) to collect information, recover data and the like. In some embodiments, anti-entropy watcher 252 may perform scans, for example, on cold index store 262, described below, and storage nodes 246, to ensure referential integrity.
[0086] In an embodiment, information stored at storage node registrar store 250 is used by a variety of services such as storage node registrar 248, storage allocator 256, anti-entropy watcher 252 and the like. For example, storage node registrar 248 may provide data location and placement services (e.g., to storage node managers 244) during data storage, retrieval and deletion. For example, given the size of a data object to be stored and information maintained by storage node registrar store 250, a storage node registrar 248 may determine where (e.g., volume) to store the data object and provides an indication of the storage location of the data object which may be used to generate a data object identifier associated with the data object. As another example, in an embodiment, storage allocator 256 uses information stored in storage node registrar store 250 to create and place volume components for new volumes in specific storage nodes to satisfy isolation and fault tolerance constraints. As yet another example, in an embodiment, anti-entropy watcher 252 uses information stored in storage node registrar store 250 to detect entropic effects such as data loss, hardware failure and the like.
[0087] In some embodiments, data plane 214 also includes an orphan cleanup data store 254, which is used to track orphans in the storage system. As used herein, an orphan is a stored data object that is not referenced by any external entity. In various embodiments, orphan cleanup data store 254 may be implemented by a NoSQL data management system, such as a key-value data store, an RDBMS or any other data storage system. In some embodiments, storage node registrars 248 stores object placement information in orphan cleanup data store 254. Subsequently, information stored in orphan cleanup data store 254 may be compared, for example, by an anti-entropy watcher 252, with information maintained in metadata plane 216. If an orphan is detected, in some embodiments, a request is inserted in the common control plane 212 to delete the orphan.
[0088] Referring now to metadata plane 216 illustrated in FIG. 2 . In various embodiments, metadata plane 216 provides information about data objects stored in the system for inventory and accounting purposes, to satisfy customer metadata inquiries and the like. In the illustrated embodiment, metadata plane 216 includes a metadata manager job store 258 which stores information about executed transactions based on entries from job result queue 242 in common control plane 212. In various embodiments, metadata manager job store 258 may be implemented by a NoSQL data management system, such as a key-value data store, a RDBMS or any other data storage system. In some embodiments, metadata manager job store 258 is partitioned and sub-partitioned, for example, based on logical data containers, to facilitate parallel processing by multiple instances of services such as metadata manager 260.
[0089] In the illustrative embodiment, metadata plane 216 also includes one or more metadata managers 260 for generating a cold index of data objects (e.g., stored in cold index store 262) based on records in metadata manager job store 258. As used herein, a “cold” index refers to an index that is updated infrequently. In various embodiments, a cold index is maintained to reduce cost overhead. In some embodiments, multiple metadata managers 260 may periodically read and process records from different partitions in metadata manager job store 258 in parallel and store the result in a cold index store 262.
[0090] In some embodiments cold index store 262 may be implemented by a reliable and durable data storage service. In some embodiments, cold index store 262 is configured to handle metadata requests initiated by customers. For example, a customer may issue a request to list all data objects contained in a given logical data container. In response to such a request, cold index store 262 may provide a list of identifiers of all data objects contained in the logical data container based on information maintained by cold index 262. In some embodiments, an operation may take a relative long period of time and the customer may be provided a job identifier to retrieve the result when the job is done. In other embodiments, cold index store 262 is configured to handle inquiries from other services, for example, from front end 208 for inventory, accounting and billing purposes.
[0091] In some embodiments, metadata plane 216 may also include a container metadata store 264 that stores information about logical data containers such as container ownership, policies, usage and the like. Such information may be used, for example, by front end 208 services, to perform authorization, metering, accounting and the like. In various embodiments, container metadata store 264 may be implemented by a NoSQL data management system, such as a key-value data store, a RDBMS or any other data storage system.
[0092] As described herein, in various embodiments, the archival data storage system 206 described herein is implemented to be efficient and scalable. For example, in an embodiment, batch processing and request coalescing is used at various stages (e.g., front end request handling, control plane job request handling, data plane data request handling) to improve efficiency. For another example, in an embodiment, processing of metadata such as jobs, requests and the like are partitioned so as to facilitate parallel processing of the partitions by multiple instances of services.
[0093] In an embodiment, data elements stored in the archival data storage system (such as data components, volumes, described below) are self-describing so as to avoid the need for a global index data structure. For example, in an embodiment, data objects stored in the system may be addressable by data object identifiers that encode storage location information. For another example, in an embodiment, volumes may store information about which data objects are stored in the volume and storage nodes and devices storing such volumes may collectively report their inventory and hardware information to provide a global view of the data stored in the system (such as evidenced by information stored in storage node registrar store 250). In such an embodiment, the global view is provided for efficiency only and not required to locate data stored in the system.
[0094] In various embodiments, the archival data storage system described herein is implemented to improve data reliability and durability. For example, in an embodiment, a data object is redundantly encoded into a plurality of data components and stored across different data storage entities to provide fault tolerance. For another example, in an embodiment, data elements have multiple levels of integrity checks. In an embodiment, parent/child relations always have additional information to ensure full referential integrity. For example, in an embodiment, bulk data transmission and storage paths are protected by having the initiator pre-calculate the digest on the data before transmission and subsequently supply the digest with the data to a receiver. The receiver of the data transmission is responsible for recalculation, comparing and then acknowledging to the sender that includes the recalculated the digest. Such data integrity checks may be implemented, for example, by front end services, transient data storage services, data plane storage entities and the like described above.
[0095] FIG. 3 illustrates an interconnection network 300 in which components of an archival data storage system may be connected, in accordance with at least one embodiment. In particular, the illustrated example shows how data plane components are connected to the interconnection network 300. In some embodiments, the interconnection network 300 may include a fat tree interconnection network where the link bandwidth grows higher or “fatter” towards the root of the tree. In the illustrated example, data plane includes one or more datacenters 301. Each datacenter 301 may include one or more storage node manager server racks 302 where each server rack hosts one or more servers that collectively provide the functionality of a storage node manager such as described in connection with FIG. 2 . In other embodiments, each storage node manager server rack may host more than one storage node manager. Configuration parameters such as number of storage node managers per rack, number of storage node manager racks and the like may be determined based on factors such as cost, scalability, redundancy and performance requirements, hardware and software resources and the like.
[0096] Each storage node manager server rack 302 may have a storage node manager rack connection 314 to an interconnect 308 used to connect to the interconnection network 300. In some embodiments, the connection 314 is implemented using a network switch 303 that may include a top-of-rack Ethernet switch or any other type of network switch. In various embodiments, interconnect 308 is used to enable high-bandwidth and low-latency bulk data transfers. For example, interconnect may include a Clos network, a fat tree interconnect, an Asynchronous Transfer Mode (ATM) network, a Fast or Gigabit Ethernet and the like.
[0097] In various embodiments, the bandwidth of storage node manager rack connection 314 may be configured to enable high-bandwidth and low-latency communications between storage node managers and storage nodes located within the same or different data centers. For example, in an embodiment, the storage node manager rack connection 314 has a bandwidth of 10 Gigabit per second (Gbps).
[0098] In some embodiments, each datacenter 301 may also include one or more storage node server racks 304 where each server rack hosts one or more servers that collectively provide the functionalities of a number of storage nodes such as described in connection with FIG. 2 . Configuration parameters such as number of storage nodes per rack, number of storage node racks, ration between storage node managers and storage nodes and the like may be determined based on factors such as cost, scalability, redundancy and performance requirements, hardware and software resources and the like. For example, in one embodiment, there are 3 storage nodes per storage node server rack, 30-80 racks per data center and a storage nodes/storage node manager ratio of 10 to 1.
[0099] Each storage node server rack 304 may have a storage node rack connection 316 to an interconnection network switch 308 used to connect to the interconnection network 300. In some embodiments, the connection 316 is implemented using a network switch 305 that may include a top-of-rack Ethernet switch or any other type of network switch. In various embodiments, the bandwidth of storage node rack connection 316 may be configured to enable high-bandwidth and low-latency communications between storage node managers and storage nodes located within the same or different data centers. In some embodiments, a storage node rack connection 316 has a higher bandwidth than a storage node manager rack connection 314. For example, in an embodiment, the storage node rack connection 316 has a bandwidth of 20 Gbps while a storage node manager rack connection 314 has a bandwidth of 10 Gbps.
[0100] In some embodiments, datacenters 301 (including storage node managers and storage nodes) communicate, via connection 310, with other computing resources services 306 such as payload data cache 228, storage node manager job store 240, storage node registrar 248, storage node registrar store 350, orphan cleanup data store 254, metadata manager job store 258 and the like as described in connection with FIG. 2 .
[0101] In some embodiments, one or more datacenters 301 may be connected via inter-datacenter connection 312. In some embodiments, connections 310 and 312 may be configured to achieve effective operations and use of hardware resources. For example, in an embodiment, connection 310 has a bandwidth of 30-100 Gbps per datacenter and inter-datacenter connection 312 has a bandwidth of 100-250 Gbps.
[0102] FIG. 4 illustrates an interconnection network 400 in which components of an archival data storage system may be connected, in accordance with at least one embodiment. In particular, the illustrated example shows how non-data plane components are connected to the interconnection network 300. As illustrated, front end services, such as described in connection with FIG. 2 , may be hosted by one or more front end server racks 402. For example, each front end server rack 402 may host one or more web servers. The front end server racks 402 may be connected to the interconnection network 400 via a network switch 408. In one embodiment, configuration parameters such as number of front end services, number of services per rack, bandwidth for front end server rack connection 314 and the like may roughly correspond to those for storage node managers as described in connection with FIG. 3 .
[0103] In some embodiments, control plane services and metadata plane services as described in connection with FIG. 2 may be hosted by one or more server racks 404. Such services may include job tracker 230, metadata manager 260, cleanup agent 234, job request balancer 238 and other services. In some embodiments, such services include services that do not handle frequent bulk data transfers. Finally, components described herein may communicate via connection 410, with other computing resources services 406 such as payload data cache 228, job tracker store 232, metadata manager job store 258 and the like as described in connection with FIG. 2 .
[0104] FIG. 5 illustrates an example process 500 for storing data, in accordance with at least one embodiment. Some or all of process 500 (or any other processes described herein or variations and/or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions and may be implemented as code (e.g., executable instructions, one or more computer programs or one or more applications) executing collectively on one or more processors, by hardware or combinations thereof. The code may be stored on a computer-readable storage medium, for example, in the form of a computer program comprising a plurality of instructions executable by one or more processors. The computer-readable storage medium may be non-transitory. In an embodiment, one or more components of archival data storage system 206 as described in connection with FIG. 2 may perform process 500.
[0105] In an embodiment, process 500 includes receiving 502 a data storage request to store archival data such as a document, a video or audio file or the like. Such a data storage request may include payload data and metadata such as size and digest of the payload data, user identification information (e.g., user name, account identifier and the like), a logical data container identifier and the like. In some embodiments, process 500 may include receiving 502 multiple storage requests each including a portion of larger payload data. In other embodiments, a storage request may include multiple data objects to be uploaded. In an embodiment, step 502 of process 500 is implemented by a service such as API request handler 218 of front end 208 as described in connection with FIG. 2 .
[0106] In an embodiment, process 500 includes processing 504 the storage request upon receiving 502 the request. Such processing may include, for example, verifying the integrity of data received, authenticating the customer, authorizing requested access against access control policies, performing meter- and accounting-related activities and the like. In an embodiment, such processing may be performed by services of front end 208 such as described in connection with FIG. 2 . In an embodiment, such a request may be processed in connection with other requests, for example, in batch mode.
[0107] In an embodiment, process 500 includes storing 506 the data associated with the storage request in a staging data store. Such staging data store may include a transient data store such as provided by payload data cache 228 as described in connection with FIG. 2 . In some embodiments, only payload data is stored in the staging store. In other embodiments, metadata related to the payload data may also be stored in the staging store. In an embodiment, data integrity is validated (e.g., based on a digest) before being stored at a staging data store.
[0108] In an embodiment, process 500 includes providing 508 a data object identifier associated with the data to be stored, for example, in a response to the storage request. As described above, a data object identifier may be used by subsequent requests to retrieve, delete or otherwise reference data stored. In an embodiment, a data object identifier may encode storage location information that may be used to locate the stored data object, payload validation information such as size, digest, timestamp and the like that may be used to validate the integrity of the payload data, metadata validation information such as error-detection codes that may be used to validate the integrity of metadata such as the data object identifier itself and information encoded in the data object identifier and the like. In an embodiment, a data object identifier may also encode information used to validate or authorize subsequent customer requests. For example, a data object identifier may encode the identifier of the logical data container that the data object is stored in. In a subsequent request to retrieve this data object, the logical data container identifier may be used to determine whether the requesting entity has access to the logical data container and hence the data objects contained therein. In some embodiments, the data object identifier may encode information based on information supplied by a customer (e.g., a global unique identifier, GUID, for the data object and the like) and/or information collected or calculated by the system performing process 500 (e.g., storage location information). In some embodiments, generating a data object identifier may include encrypting some or all of the information described above using a cryptographic private key. In some embodiments, the cryptographic private key may be periodically rotated. In some embodiments, a data object identifier may be generated and/or provided at a different time than described above. For example, a data object identifier may be generated and/or provided after a storage job (described below) is created and/or completed.
[0109] In an embodiment, providing 508 a data object identifier may include determining a storage location for the before the data is actually stored there. For example, such determination may be based at least in part on inventory information about existing data storage entities such as operational status (e.g., active or inactive), available storage space, data isolation requirement and the like. In an environment such as environment 200 illustrated by FIG. 2 , such determination may be implemented by a service such as storage node registrar 248 as described above in connection with FIG. 2 . In some embodiments, such determination may include allocating new storage space (e.g., volume) on one or more physical storage devices by a service such as storage allocator 256 as described in connection with FIG. 2 .
[0110] In an embodiment, a storage location identifier may be generated to represent the storage location determined above. Such a storage location identifier may include, for example, a volume reference object which comprises a volume identifier component and data object identifier component. The volume reference component may identify the volume the data is stored on and the data object identifier component may identify where in the volume the data is stored. In general, the storage location identifier may comprise components that identify various levels within a logical or physical data storage topology (such as a hierarchy) in which data is organized. In some embodiments, the storage location identifier may point to where actual payload data is stored or a chain of reference to where the data is stored.
[0111] In an embodiments, a data object identifier encodes a digest (e.g., a hash) of at least a portion of the data to be stored, such as the payload data. In some embodiments, the digest may be based at least in part on a customer-provided digest. In other embodiments, the digest may be calculated from scratch based on the payload data.
[0112] In an embodiment, process 500 includes creating 510 a storage job for persisting data to a long-term data store and scheduling 512 the storage job for execution. In environment 200 as described in connection with FIG. 2 , steps 508, 510 and 512 may be implemented at least in part by components of control plane for direct I/O 210 and common control plane 212 as described above. Specifically, in an embodiment, job tracker 230 creates a job record and stores the job record in job tracker store 232. As described above, job tracker 230 may perform batch processing to reduce the total number of transactions against job tracker store 232. Additionally, job tracker store 232 may be partitioned or otherwise optimized to facilitate parallel processing, cleanup operations and the like. A job record, as described above, may include job-related information such as a customer account identifier, job identifier, storage location identifier, reference to data stored in payload data cache 228, job status, job creation and/or expiration time and the like. In some embodiments, a storage job may be created before a data object identifier is generated and/or provided. For example, a storage job identifier, instead of or in addition to a data object identifier, may be provided in response to a storage request at step 508 above.
[0113] In an embodiment, scheduling 512 the storage job for execution includes performing job planning and optimization, such as queue-based load leveling or balancing, job partitioning and the like, as described in connection with common control plane 212 of FIG. 2 . For example, in an embodiment, job request balancer 238 transfers job items from job request queue 236 to storage node manager job store 240 according to a scheduling algorithm so as to dampen peak to average load levels (jobs) coming from control plane for I/O 210 and to deliver manageable workload to data plane 214. As another example, storage node manager job store 240 may be partitioned to facilitate parallel processing of the jobs by multiple workers such as storage node managers 244. As yet another example, storage node manager job store 240 may provide querying, sorting and other functionalities to facilitate batch processing and other job optimizations.
[0114] In an embodiment, process 500 includes selecting 514 the storage job for execution, for example, by a storage node manager 244 from storage node manager job stored 240 as described in connection with FIG. 2 . The storage job may be selected 514 with other jobs for batch processing or otherwise selected as a result of job planning and optimization described above.
[0115] In an embodiment, process 500 includes obtaining 516 data from a staging store, such as payload data cache 228 described above in connection with FIG. 2 . In some embodiments, the integrity of the data may be checked, for example, by verifying the size, digest, an error-detection code and the like.
[0116] In an embodiment, process 500 includes obtaining 518 one or more data encoding schemes such as an encryption scheme, a redundancy encoding scheme such as erasure encoding, redundant array of independent disks (RAID) encoding schemes, replication, and the like. In some embodiments, such encoding schemes evolve to adapt to different requirements. For example, encryption keys may be rotated periodically and stretch factor of an erasure coding scheme may be adjusted over time to different hardware configurations, redundancy requirements and the like.
[0117] In an embodiment, process 500 includes encoding 520 with the obtained encoding schemes. For example, in an embodiment, data is encrypted and the encrypted data is erasure-encoded. In an embodiment, storage node managers 244 described in connection with FIG. 2 may be configured to perform the data encoding described herein. In an embodiment, application of such encoding schemes generates a plurality of encoded data components or shards, which may be stored across different storage entities such as storage devices, storage nodes, datacenters and the like to provide fault tolerance. In an embodiment where data may comprise multiple parts (such as in the case of a multi-part upload), each part may be encoded and stored as described herein.
[0118] In an embodiment, process 500 includes determining 522 the storage entities for such encoded data components. For example, in an environment 200 illustrated by FIG. 2 , a storage node manager 244 may determine the plurality of storage nodes 246 to store the encoded data components by querying a storage node registrar 248 using a volume identifier. Such a volume identifier may be part of a storage location identifier associated with the data to be stored. In response to the query with a given volume identifier, in an embodiment, storage node registrar 248 returns a list of network locations (including endpoints, DNS names, IP addresses and the like) of storage nodes 246 to store the encoded data components. As described in connection with FIG. 2 , storage node registrar 248 may determine such a list based on self-reported and dynamically provided and/or updated inventory information from storage nodes 246 themselves. In some embodiments, such determination is based on data isolation, fault tolerance, load balancing, power conservation, data locality and other considerations. In some embodiments, storage registrar 248 may cause new storage space to be allocated, for example, by invoking storage allocator 256 as described in connection with FIG. 2 .
[0119] In an embodiment, process 500 includes causing 524 storage of the encoded data component(s) at the determined storage entities. For example, in an environment 200 illustrated by FIG. 2 , a storage node manager 244 may request each of the storage nodes 246 determined above to store a data component at a given storage location. Each of the storage nodes 246, upon receiving the storage request from storage node manager 244 to store a data component, may cause the data component to be stored in a connected storage device. In some embodiments, at least a portion of the data object identifier is stored with all or some of the data components in either encoded or unencoded form. For example, the data object identifier may be stored in the header of each data component and/or in a volume component index stored in a volume component. In some embodiments, a storage node 246 may perform batch processing or other optimizations to process requests from storage node managers 244.
[0120] In an embodiment, a storage node 246 sends an acknowledgement to the requesting storage node manager 244 indicating whether data is stored successfully. In some embodiments, a storage node 246 returns an error message, when for some reason, the request cannot be fulfilled. For example, if a storage node receives two requests to store to the same storage location, one or both requests may fail. In an embodiment, a storage node 246 performs validation checks prior to storing the data and returns an error if the validation checks fail. For example, data integrity may be verified by checking an error-detection code or a digest. As another example, storage node 246 may verify, for example, based on a volume index, that the volume identified by a storage request is stored by the storage node and/or that the volume has sufficient space to store the data component.
[0121] In some embodiments, data storage is considered successful when storage node manager 244 receives positive acknowledgement from at least a subset (a storage quorum) of requested storage nodes 246. In some embodiments, a storage node manager 244 may wait until the receipt of a quorum of acknowledgement before removing the state necessary to retry the job. Such state information may include encoded data components for which an acknowledgement has not been received. In other embodiments, to improve the throughput, a storage node manager 244 may remove the state necessary to retry the job before receiving a quorum of acknowledgement
[0122] In an embodiment, process 500 includes updating 526 metadata information including, for example, metadata maintained by data plane 214 (such as index and storage space information for a storage device, mapping information stored at storage node registrar store 250 and the like), metadata maintained by control planes 210 and 212 (such as job-related information), metadata maintained by metadata plane 216 (such as a cold index) and the like. In various embodiments, some of such metadata information may be updated via batch processing and/or on a periodic basis to reduce performance and cost impact. For example, in data plane 214, information maintained by storage node registrar store 250 may be updated to provide additional mapping of the volume identifier of the newly stored data and the storage nodes 246 on which the data components are stored, if such a mapping is not already there. For another example, volume index on storage devices may be updated to reflect newly added data components.
[0123] In common control plane 212, job entries for completed jobs may be removed from storage node manager job store 240 and added to job result queue 242 as described in connection with FIG. 2 . In control plane for direct I/O 210, statuses of job records in job tracker store 232 may be updated, for example, by job tracker 230 which monitors the job result queue 242. In various embodiments, a job that fails to complete may be retried for a number of times. For example, in an embodiment, a new job may be created to store the data at a different location. As another example, an existing job record (e.g., in storage node manager job store 240, job tracker store 232 and the like) may be updated to facilitate retry of the same job.
[0124] In metadata plane 216, metadata may be updated to reflect the newly stored data. For example, completed jobs may be pulled from job result queue 242 into metadata manager job store 258 and batch-processed by metadata manager 260 to generate an updated index such as stored in cold index store 262. For another example, customer information may be updated to reflect changes for metering and accounting purposes.
[0125] Finally, in some embodiments, once a storage job is completed successfully, job records, payload data and other data associated with a storage job may be removed, for example, by a cleanup agent 234 as described in connection with FIG. 2 . In some embodiments, such removal may be processed by batch processing, parallel processing or the like.
[0126] FIG. 6 illustrates an example process 500 for retrieving data, in accordance with at least one embodiment. In an embodiment, one or more components of archival data storage system 206 as described in connection with FIG. 2 collectively perform process 600.
[0127] In an embodiment, process 600 includes receiving 602 a data retrieval request to retrieve data such as stored by process 500, described above. Such a data retrieval request may include a data object identifier, such as provided by step 508 of process 500, described above, or any other information that may be used to identify the data to be retrieved.
[0128] In an embodiment, process 600 includes processing 604 the data retrieval request upon receiving 602 the request. Such processing may include, for example, authenticating the customer, authorizing requested access against access control policies, performing meter and accounting related activities and the like. In an embodiment, such processing may be performed by services of front end 208 such as described in connection with FIG. 2 . In an embodiment, such request may be processed in connection with other requests, for example, in batch mode.
[0129] In an embodiment, processing 604 the retrieval request may be based at least in part on the data object identifier that is included in the retrieval request. As described above, data object identifier may encode storage location information, payload validation information such as size, creation timestamp, payload digest and the like, metadata validation information, policy information and the like. In an embodiment, processing 604 the retrieval request includes decoding the information encoded in the data object identifier, for example, using a private cryptographic key and using at least some of the decoded information to validate the retrieval request. For example, policy information may include access control information that may be used to validate that the requesting entity of the retrieval request has the required permission to perform the requested access. As another example, metadata validation information may include an error-detection code such as a cyclic redundancy check (“CRC”) that may be used to verify the integrity of data object identifier or a component of it.
[0130] In an embodiment, process 600 includes creating 606 a data retrieval job corresponding to the data retrieval request and providing 608 a job identifier associated with the data retrieval job, for example, in a response to the data retrieval request. In some embodiments, creating 606 a data retrieval job is similar to creating a data storage job as described in connection with step 510 of process 500 illustrated in FIG. 5 . For example, in an embodiment, a job tracker 230 may create a job record that includes at least some information encoded in the data object identifier and/or additional information such as a job expiration time and the like and store the job record in job tracker store 232. As described above, job tracker 230 may perform batch processing to reduce the total number of transactions against job tracker store 232. Additionally, job tracker store 232 may be partitioned or otherwise optimized to facilitate parallel processing, cleanup operations and the like.
[0131] In an embodiment, process 600 includes scheduling 610 the data retrieval job created above. In some embodiments, scheduling 610 the data retrieval job for execution includes performing job planning and optimization such as described in connection with step 512 of process 500 of FIG. 5 . For example, the data retrieval job may be submitted into a job queue and scheduled for batch processing with other jobs based at least in part on costs, power management schedules and the like. For another example, the data retrieval job may be coalesced with other retrieval jobs based on data locality and the like.
[0132] In an embodiment, process 600 includes selecting 612 the data retrieval job for execution, for example, by a storage node manager 244 from storage node manager job stored 240 as described in connection with FIG. 2 . The retrieval job may be selected 612 with other jobs for batch processing or otherwise selected as a result of job planning and optimization described above.
[0133] In an embodiment, process 600 includes determining 614 the storage entities that store the encoded data components that are generated by a storage process such as process 500 described above. In an embodiment, a storage node manager 244 may determine a plurality of storage nodes 246 to retrieve the encoded data components in a manner similar to that discussed in connection with step 522 of process 500, above. For example, such determination may be based on load balancing, power conservation, efficiency and other considerations.
[0134] In an embodiment, process 600 includes determining 616 one or more data decoding schemes that may be used to decode retrieved data. Typically, such decoding schemes correspond to the encoding schemes applied to the original data when the original data is previously stored. For example, such decoding schemes may include decryption with a cryptographic key, erasure-decoding and the like.
[0135] In an embodiment, process 600 includes causing 618 retrieval of at least some of the encoded data components from the storage entities determined in step 614 of process 600. For example, in an environment 200 illustrated by FIG. 2 , a storage node manager 244 responsible for the data retrieval job may request a subset of storage nodes 246 determined above to retrieve their corresponding data components. In some embodiments, a minimum number of encoded data components is needed to reconstruct the original data where the number may be determined based at least in part on the data redundancy scheme used to encode the data (e.g., stretch factor of an erasure coding). In such embodiments, the subset of storage nodes may be selected such that no less than the minimum number of encoded data components is retrieved.
[0136] Each of the subset of storage nodes 246, upon receiving a request from storage node manager 244 to retrieve a data component, may validate the request, for example, by checking the integrity of a storage location identifier (that is part of the data object identifier), verifying that the storage node indeed holds the requested data component and the like. Upon a successful validation, the storage node may locate the data component based at least in part on the storage location identifier. For example, as described above, the storage location identifier may include a volume reference object which comprises a volume identifier component and a data object identifier component where the volume reference component to identify the volume the data is stored and a data object identifier component may identify where in the volume the data is stored. In an embodiment, the storage node reads the data component, for example, from a connected data storage device and sends the retrieved data component to the storage node manager that requested the retrieval. In some embodiments, the data integrity is checked, for example, by verifying the data component identifier or a portion thereof is identical to that indicated by the data component identifier associated with the retrieval job. In some embodiments, a storage node may perform batching or other job optimization in connection with retrieval of a data component.
[0137] In an embodiment, process 600 includes decoding 620, at least the minimum number of the retrieved encoded data components with the one or more data decoding schemes determined at step 616 of process 600. For example, in one embodiment, the retrieved data components may be erasure decoded and then decrypted. In some embodiments, a data integrity check is performed on the reconstructed data, for example, using payload integrity validation information encoded in the data object identifier (e.g., size, timestamp, digest). In some cases, the retrieval job may fail due to a less-than-minimum number of retrieved data components, failure of data integrity check and the like. In such cases, the retrieval job may be retried in a fashion similar to that described in connection with FIG. 5 . In some embodiments, the original data comprises multiple parts of data and each part is encoded and stored. In such embodiments, during retrieval, the encoded data components for each part of the data may be retrieved and decoded (e.g., erasure-decoded and decrypted) to form the original part and the decoded parts may be combined to form the original data.
[0138] In an embodiment, process 600 includes storing reconstructed data in a staging store such as payload data cache 228 described in connection with FIG. 2 . In some embodiments, data stored 622 in the staging store may be available for download by a customer for a period of time or indefinitely. In an embodiment, data integrity may be checked (e.g., using a digest) before the data is stored in the staging store.
[0139] In an embodiment, process 600 includes providing 624 a notification of the completion of the retrieval job to the requestor of the retrieval request or another entity or entities otherwise configured to receive such a notification. Such notifications may be provided individually or in batches. In other embodiments, the status of the retrieval job may be provided upon a polling request, for example, from a customer.
[0140] FIG. 7 illustrates an example process 700 for deleting data, in accordance with at least one embodiment. In an embodiment, one or more components of archival data storage system 206 as described in connection with FIG. 2 collectively perform process 700.
[0141] In an embodiment, process 700 includes receiving 702 a data deletion request to delete data such as stored by process 500, described above. Such a data retrieval request may include a data object identifier, such as provided by step 508 of process 500, described above, or any other information that may be used to identify the data to be deleted.
[0142] In an embodiment, process 700 includes processing 704 the data deletion request upon receiving 702 the request. In some embodiments, the processing 704 is similar to that for step 504 of process 500 and step 604 of process 600, described above. For example, in an embodiment, the processing 704 is based at least in part on the data object identifier that is included in the data deletion request.
[0143] In an embodiment, process 700 includes creating 706 a data retrieval job corresponding to the data deletion request. Such a retrieval job may be created similar to the creation of storage job described in connection with step 510 of process 500 and the creation of the retrieval job described in connection with step 606 of process 600.
[0144] In an embodiment, process 700 includes providing 708 an acknowledgement that the data is deleted. In some embodiments, such acknowledgement may be provided in response to the data deletion request so as to provide a perception that the data deletion request is handled synchronously. In other embodiments, a job identifier associated with the data deletion job may be provided similar to the providing of job identifiers for data retrieval requests.
[0145] In an embodiment, process 700 includes scheduling 708 the data deletion job for execution. In some embodiments, scheduling 708 of data deletion jobs may be implemented similar to that described in connection with step 512 of process 500 and in connection with step 610 of process 600, described above. For example, data deletion jobs for closely-located data may be coalesced and/or batch processed. For another example, data deletion jobs may be assigned a lower priority than data retrieval jobs.
[0146] In some embodiments, data stored may have an associated expiration time that is specified by a customer or set by default. In such embodiments, a deletion job may be created 706 and schedule 710 automatically on or near the expiration time of the data. In some embodiments, the expiration time may be further associated with a grace period during which data is still available or recoverable. In some embodiments, a notification of the pending deletion may be provided before, on or after the expiration time.
[0147] In some embodiments, process 700 includes selecting 712 the data deletion job for execution, for example, by a storage node manager 244 from storage node manager job stored 240 as described in connection with FIG. 2 . The deletion job may be selected 712 with other jobs for batch processing or otherwise selected as a result of job planning and optimization described above.
[0148] In some embodiments, process 700 includes determining 714 the storage entities for data components that store the data components that are generated by a storage process such as process 500 described above. In an embodiment, a storage node manager 244 may determine a plurality of storage nodes 246 to retrieve the encoded data components in a manner similar to that discussed in connection with step 614 of process 600 described above.
[0149] In some embodiments, process 700 includes causing 716 the deletion of at least some of the data components. For example, in an environment 200 illustrated by FIG. 2 , a storage node manager 244 responsible for the data deletion job may identify a set of storage nodes that store the data components for the data to be deleted and requests at least a subset of those storage nodes to delete their respective data components. Each of the subset of storage node 246, upon receiving a request from storage node manager 244 to delete a data component, may validate the request, for example, by checking the integrity of a storage location identifier (that is part of the data object identifier), verifying that the storage node indeed holds the requested data component and the like. Upon a successful validation, the storage node may delete the data component from a connected storage device and sends an acknowledgement to storage node manager 244 indicating whether the operation was successful. In an embodiment, multiple data deletion jobs may be executed in a batch such that data objects located close together may be deleted as a whole. In some embodiments, data deletion is considered successful when storage node manager 244 receives positive acknowledgement from at least a subset of storage nodes 246. The size of the subset may be configured to ensure that data cannot be reconstructed later on from undeleted data components. Failed or incomplete data deletion jobs may be retried in a manner similar to the retrying of data storage jobs and data retrieval jobs, described in connection with process 500 and process 600, respectively.
[0150] In an embodiment, process 700 includes updating 718 metadata information such as that described in connection with step 526 of process 500. For example, storage nodes executing the deletion operation may update storage information including index, free space information and the like. In an embodiment, storage nodes may provide updates to storage node registrar or storage node registrar store. In various embodiments, some of such metadata information may be updated via batch processing and/or on a periodic basis to reduce performance and cost impact.
[0151] In some embodiments, some or all of the storage devices associated with the storage nodes may utilize, for example, shingled magnetic recording (SMR) techniques. FIG. 8 provides an illustrative example of a storage media layout 800 for storage devices using SMR techniques. The example illustrated shows three data tracks A 802, B 804 and C 806, overlaid atop each other in an overlapping, or “shingled” configuration. Each track consists of multiple data locations that are written in sequence, or “down-track.” Once one track has fully been written, the head traverses to the next track, or “down-shingle,” and proceeds to write each associated data location within that track in sequence as well. This process continues sequentially. As noted by the figure, the effective width of a magnetic write head 808 may correspond to the full width of a data track. In other words, the width, measured radially, of the effect of a magnetic write head upon data locations passing underneath the head may be as wide as a data track if it was unobstructed by other tracks. However, since the tracks overlap due to the shingled configuration, the effective track width 810 is less than that of the write head. The head is typically positioned such that the radially furthest outward effect upon data locations is aligned to the furthest up-shingle portion of a given track. Thus, so aligned and due to the effective width of the head, it may be contemplated that a write to a given track will also affect the data locations of one or more tracks down-shingle from the given track. When down-shingle tracks contain no data, this effect is of little concern. Thus, media using SMR techniques are preferably written in sequential fashion, from start to finish, so that tracks down-shingle from a track being written remain unwritten. Such media characteristics may warrant specific routines and schemes, such as the log-based storage scheme described herein, in order to prevent write fragmentation, maximize write and read efficiency and improve data storage reliability, as, for example, would be desirable in an archival data storage system as previously described in connection with at least FIGS. 2-7 .
[0152] FIG. 9 provides an illustrative example of various components of a log-based storage scheme 900. As previously described, in some embodiments, the overall log-based storage scheme is sequentially written such that records precede and succeed each group of data, which in turn may contain one or more subparts of data. In the example provided, a generational record 902 is written to the “top” of the media (i.e., the beginning) so as to reflect an initialization of the media. Such a record may include one or more of a device ID that may be specific to the drive or a connecting system or entity, a unique sequence number to identify the storage device or media upon which the data are or will be stored, or other information that allows a connecting entity (or the storage device itself) to disambiguate data stored upon the device. A connecting entity may read or detect the generational record to, for example, determine whether a drive has been initialized and is thus ready to accept data.
[0153] In some embodiments, the generational record is followed by one or more data sets 904. Each data set may contain, sequentially in some embodiments, a record of intent 906, a record of information 908 containing one or more page records 910 (some of which may, in the course of operation, be deleted or unsuccessfully written 912) and a record of seal 914. In some embodiments, the record of intent contains information about the data to follow, such as the data written within the record of information. The information may include (but is not limited to) one or more elements of the generational record, such as the device ID or device sequence number, information identifying the record (such as a record sequence identifier) or one or more portions of a record of information to follow, information regarding the intended size and/or contents of data to be written, or any other information that a connecting entity (such as a data storage node of the archival data storage system described above) or the storage device would appropriately place in such a record.
[0154] In some embodiments, a record of information 908 follows the record of intent 906. The record of information may, in some embodiments, contain at least a record or device sequence identifier, which may be similar to those included in the generational record and/or record of intent, as well as the raw data intended to be written and as anticipated by the record of intent. The contained data may be broken into any number of portions, such as one or more page records 910. The page records may vary in size, or in some embodiments, may be of a uniform and/or predetermined size. Such page records may also include a record or device sequence identifier and/or integrity data associated with the raw data contained within the page record. The integrity data may, in some embodiments, be generated by the data storage device or an entity or system connected therewith, by applying one or more algorithms (such as erasure codes, hash functions and the like) to the raw data within the page record. In some embodiments, aborted writes of raw data and/or page records 912, such as those induced by faulty data locations upon the media, are skipped such that subsequent data writes (e.g., of the raw data or further page records) are not unnecessarily or indefinitely forestalled.
[0155] In some embodiments, a record of seal 914 sequentially follows the record of information. The record of seal may serve as a “catalog” of information written within the record of information, such as the constituent page records and associated stored data. For example, the record of seal may provide specific locations for portions of data within the record of information that were successfully written. In some embodiments, any failed data locations, aborted writes or otherwise unsuccessfully stored page records would be either be omitted or specifically noted as unavailable. It is contemplated that the record of seal may contain references to data within other data sets. For example, if a deletion request for a given portion of data is received and/or processed after a page record for a given data set has been committed to the drive, an entry or record noting the unavailability of the associated data may be written within the record of seal for the data set being written at the current location. The record of seal may also include information that ties together one or more page records, either within the record of information for the associated data set or for those that precedes it, as a volume component that in turn may be a constituent of a logical volume.
[0156] As the record of intent defines, in some embodiments, an expected size of the record of information or the data contained within it, such information may be used by a connecting entity to, upon reading such information from the record of intent, to quickly locate the associated record of seal at the end of the record of information. As a record of seal is immediately followed, in some embodiments, by a record of intent of the sequentially next data set, a connected entity may traverse from one data set to another. A connected entity may do so in response to one or more read requests (e.g., to locate a specific data object or portion), preemptively (e.g., to retain a current state of the storage device's stored data so that the retained state may be searched for a given data object or portion without needing to traverse the storage device itself), or some combination of the two. For example, various approaches may be used to improve the efficiency of a read request, such as by only reading metadata associated with each record of intent or record of seal rather than the entire associated indexes, and only fully processing the indexes and traversing them in response to a read request for specific data referred to by the metadata. As another example, the state of the drive may be preemptively generated, then subsequently only refreshed by a partial traversal of the records upon notification that certain records have been updated.
[0157] In some embodiments, a checkpoint record containing aggregated indexes 916 may be placed at predetermined and/or fixed locations upon the storage device. For example, an aggregated checkpoint record may be located at every 1 TB of a storage device. As another example, such a record may be placed after every N data sets. Such aggregated checkpoint records may contain an index of some or all of the data stored sequentially prior to the checkpoint record. Thus, when a connected entity requires data stored upon the storage device to be located, it need only cause the storage device to seek directly to, e.g., the sequentially final aggregated checkpoint record and locate the appropriate reference therein. When reading the storage device to determine or generate a representation of its state, the connected entity may similar seek directly to the sequentially final checkpoint record, process the indexes contained therein, then traverse and process the records located thereafter using techniques previously described until either the end of the drive or the location of a write pointer that indicates a final current write position on the storage device.
[0158] FIG. 10 illustrates an example process 1000 for queuing and/or caching data for storage in accordance with some embodiments. Some or all of the process 1000 (or any other processes described herein, or variations and/or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions and may be implemented as code (e.g., executable instructions, one or more computer programs or one or more applications) executing collectively on one or more processors, by hardware or combinations thereof. The code may be stored on a computer-readable storage medium, for example, in the form of a computer program comprising a plurality of instructions executable by one or more processors. The computer-readable storage medium may be non-transitory. Some or all of process 1000 may be performed, for example, by various components of the archival data storage system 206 described in connection with at least FIGS. 2-7 or the queuing and/or caching techniques illustrated in connection with FIG. 1 , such as the storage node manager 244, data storage nodes 246 and/or 102 and/or data storage devices 106 and associated controllers and/or adapters 110.
[0159] One or more data storage devices that sequentially store data are selected 1002, based, in some embodiments, at least in part on their ability to respond to write pointer queries. In some embodiments, such data storage devices may be adapted and/or modified, for example by updating firmware or by programming a connected controller, to externally provide information relating to the write pointer, as described at least in connection with FIG. 1 . Data storage devices may be selected, for example, based at least in part on information about various operational parameters of the data storage device, such as its current operational state, available storage space, locality requirements and the like. In some embodiments, the storage node registrar 248 and/or the storage allocator 256 may make a determination of which data storage devices among a plurality of data storage devices connected to, for example, the archival data storage system 206 meet predefined operational criteria, such as those relating to the operational parameters just described. In some embodiments, at least a subset of the determined data storage devices are selected to store the data. Data write requests are received 1004, e.g., from connected entities such as one or more of the various components of the archival data storage system, and processed such that data associated with the data write requests are cached and/or queued 1006. Data write requests may be received in association with receiving a data storage request to store data, such as from a customer or user of the archival data storage system. In some embodiments, data write requests may be received as a result of some autonomous process of an archival data storage system, such as during system maintenance, internal data consolidation, garbage collection routines (e.g., orphan removal as described in connection with at least FIG. 2 ) and the like.
[0160] As previously described, the data and/or associated data write requests may be queued by any appropriate entity, such as a data storage node. The data may be cached and/or queued in any appropriate order, including an order that is not necessarily related to the order in which the associated data write requests were received. The cached data are stored upon the selected data storage device(s) in a specified sequence 1008, which, as previously mentioned, may be optimized for performance and/or efficiency of either the data storage device(s) or the connected entities. Information relating to the write pointer is requested and/or sent 1010, and the received information is compared with the contents and/or current position within the data queue or cache 1012 in order to, in some embodiments, determine what portion of the data elements have been committed (and/or the identity of the specific data elements committed or yet to be committed). The write pointer may be requested and/or sent any number of times over a given time period, such as upon a preset schedule and/or based on detection of operational parameters of, for example, the data storage node, the data storage device and/or the device requesting and/or receiving the information relating to the write pointer. The write pointer may be requested and/or sent to by, for example, storage node managers 244, storage nodes 246 and/or 102, payload data cache 228 and the like, so as to determine the write state of the data storage device and/or the associated data. As previously described at least in connection with FIG. 1 , the comparison may, in some embodiments, be used in lieu of cache and/or queue flushing techniques. Also as previously described, the write pointer may be implemented, in some embodiments, such that the request for the write pointer and/or the servicing of such a request may be executed such that other operations of the system, data storage device and/or other connected entities are minimally affected (or such effects are at least reduced relative to, for example, scanning the media of the data storage device to determine where the last write position is located).
[0161] FIG. 11 illustrates an example process 1100 for handling storage requests in accordance with some embodiments. One or more storage requests are received 1102. In some embodiments, the storage requests are related to storage jobs received by one or more components of the archival data storage system 206 upstream of, for example, the storage node managers 244 and/or the storage nodes 246 and/or 102. As noted at least in connection with FIGS. 2 and 5 , in some embodiments, the storage node managers may first determine an order and/or destination storage node by which to service the storage requests 1106. In some embodiments, the storage node managers may then effect the servicing of the storage requests 1106 by, for example, requesting that each of the storage nodes, to which one or more data storage devices are at least connected, store a given data component at a given storage location amongst the one or more data storage devices. As previously noted in connection with at least FIGS. 1 and 10 , such storage requests are queued at the storage nodes, which themselves may determine an order in which to service the requests. In some embodiments, the data storage manager determines, from results of the storage requests received from the storage nodes, whether any storage requests need to be retried 1110. As described in connection with at least FIG. 5 , in some embodiments, data storage is considered successful when the storage node manager receives positive acknowledgement from at least a subset of the requested storage nodes and will, in some embodiments, wait to make the determination of success or failure until receiving a quorum of acknowledgement before removing the state necessary to retry the associated job. Thus, improved efficiency of determining whether a given data request has been successfully processed, such as by implementing the write pointer-based acknowledgement scheme described in connection with at least FIGS. 1 and 10 , would increase throughput of the storage nodes, the storage node manager and various other aspects of the archival data storage system. If such a quorum of acknowledgement is not achieved, however, the failed storage requests may be retried 1112 using one or more steps among 1102-1110 of process 1100.
[0162] FIG. 12 illustrates aspects of an example environment 1200 for implementing aspects in accordance with various embodiments. As will be appreciated, although a Web-based environment is used for purposes of explanation, different environments may be used, as appropriate, to implement various embodiments. The environment includes an electronic client device 1202, which can include any appropriate device operable to send and receive requests, messages or information over an appropriate network 1204 and convey information back to a user of the device. Examples of such client devices include personal computers, cell phones, handheld messaging devices, laptop computers, set-top boxes, personal data assistants, electronic book readers and the like. The network can include any appropriate network, including an intranet, the Internet, a cellular network, a local area network or any other such network or combination thereof. Components used for such a system can depend at least in part upon the type of network and/or environment selected. Protocols and components for communicating via such a network are well known and will not be discussed herein in detail. Communication over the network can be enabled by wired or wireless connections and combinations thereof. In this example, the network includes the Internet, as the environment includes a Web server 1206 for receiving requests and serving content in response thereto, although for other networks an alternative device serving a similar purpose could be used as would be apparent to one of ordinary skill in the art.
[0163] The illustrative environment includes at least one application server 1208 and a data store 1210. It should be understood that there can be several application servers, layers, or other elements, processes or components, which may be chained or otherwise configured, which can interact to perform tasks such as obtaining data from an appropriate data store. As used herein the term “data store” refers to any device or combination of devices capable of storing, accessing and retrieving data, which may include any combination and number of data servers, databases, data storage devices and data storage media, in any standard, distributed or clustered environment. The application server can include any appropriate hardware and software for integrating with the data store as needed to execute aspects of one or more applications for the client device, handling a majority of the data access and business logic for an application. The application server provides access control services in cooperation with the data store, and is able to generate content such as text, graphics, audio and/or video to be transferred to the user, which may be served to the user by the Web server in the form of HTML, XML or another appropriate structured language in this example. The handling of all requests and responses, as well as the delivery of content between the client device 1202 and the application server 1208, can be handled by the Web server. It should be understood that the Web and application servers are not required and are merely example components, as structured code discussed herein can be executed on any appropriate device or host machine as discussed elsewhere herein.
[0164] The data store 1210 can include several separate data tables, databases or other data storage mechanisms and media for storing data relating to a particular aspect. For example, the data store illustrated includes mechanisms for storing production data 1212 and user information 1216, which can be used to serve content for the production side. The data store also is shown to include a mechanism for storing log data 1214, which can be used for reporting, analysis or other such purposes. It should be understood that there can be many other aspects that may need to be stored in the data store, such as for page image information and to access right information, which can be stored in any of the above listed mechanisms as appropriate or in additional mechanisms in the data store 1210. The data store 1210 is operable, through logic associated therewith, to receive instructions from the application server 1208 and obtain, update or otherwise process data in response thereto. In one example, a user might submit a search request for a certain type of item. In this case, the data store might access the user information to verify the identity of the user, and can access the catalog detail information to obtain information about items of that type. The information then can be returned to the user, such as in a results listing on a Web page that the user is able to view via a browser on the user device 1202. Information for a particular item of interest can be viewed in a dedicated page or window of the browser.
[0165] Each server typically will include an operating system that provides executable program instructions for the general administration and operation of that server, and typically will include a computer-readable storage medium (e.g., a hard disk, random access memory, read only memory, etc.) storing instructions that, when executed by a processor of the server, allow the server to perform its intended functions. Suitable implementations for the operating system and general functionality of the servers are known or commercially available, and are readily implemented by persons having ordinary skill in the art, particularly in light of the disclosure herein.
[0166] The environment in one embodiment is a distributed computing environment utilizing several computer systems and components that are interconnected via communication links, using one or more computer networks or direct connections. However, it will be appreciated by those of ordinary skill in the art that such a system could operate equally well in a system having fewer or a greater number of components than are illustrated in FIG. 12 . Thus, the depiction of the system 1200 in FIG. 12 should be taken as being illustrative in nature, and not limiting to the scope of the disclosure.
[0167] The various embodiments further can be implemented in a wide variety of operating environments, which in some cases can include one or more user computers, computing devices or processing devices which can be used to operate any of a number of applications. User or client devices can include any of a number of general purpose personal computers, such as desktop or laptop computers running a standard operating system, as well as cellular, wireless and handheld devices running mobile software and capable of supporting a number of networking and messaging protocols. Such a system also can include a number of workstations running any of a variety of commercially-available operating systems and other known applications for purposes such as development and database management. These devices also can include other electronic devices, such as dummy terminals, thin-clients, gaming systems and other devices capable of communicating via a network.
[0168] Most embodiments utilize at least one network that would be familiar to those skilled in the art for supporting communications using any of a variety of commercially-available protocols, such as TCP/IP, OSI, FTP, UPnP, NFS, CIFS and AppleTalk. The network can be, for example, a local area network, a wide-area network, a virtual private network, the Internet, an intranet, an extranet, a public switched telephone network, an infrared network, a wireless network and any combination thereof.
[0169] In embodiments utilizing a Web server, the Web server can run any of a variety of server or mid-tier applications, including HTTP servers, FTP servers, CGI servers, data servers, Java servers and business application servers. The server(s) also may be capable of executing programs or scripts in response requests from user devices, such as by executing one or more Web applications that may be implemented as one or more scripts or programs written in any programming language, such as Java®, C, C# or C++, or any scripting language, such as Perl, Python or TCL, as well as combinations thereof. The server(s) may also include database servers, including without limitation those commercially available from Oracle®, Microsoft®, Sybase® and IBM®.
[0170] The environment can include a variety of data stores and other memory and storage media as discussed above. These can reside in a variety of locations, such as on a storage medium local to (and/or resident in) one or more of the computers or remote from any or all of the computers across the network. In a particular set of embodiments, the information may reside in a storage-area network (“SAN”) familiar to those skilled in the art. Similarly, any necessary files for performing the functions attributed to the computers, servers or other network devices may be stored locally and/or remotely, as appropriate. Where a system includes computerized devices, each such device can include hardware elements that may be electrically coupled via a bus, the elements including, for example, at least one central processing unit (CPU), at least one input device (e.g., a mouse, keyboard, controller, touch screen or keypad), and at least one output device (e.g., a display device, printer or speaker). Such a system may also include one or more storage devices, such as disk drives, optical storage devices, and solid-state storage devices such as random access memory (“RAM”) or read-only memory (“ROM”), as well as removable media devices, memory cards, flash cards, etc.
[0171] Such devices also can include a computer-readable storage media reader, a communications device (e.g., a modem, a network card (wireless or wired), an infrared communication device, etc.) and working memory as described above. The computer-readable storage media reader can be connected with, or configured to receive, a computer-readable storage medium, representing remote, local, fixed and/or removable storage devices as well as storage media for temporarily and/or more permanently containing, storing, transmitting and retrieving computer-readable information. The system and various devices also typically will include a number of software applications, modules, services or other elements located within at least one working memory device, including an operating system and application programs, such as a client application or Web browser. It should be appreciated that alternate embodiments may have numerous variations from that described above. For example, customized hardware might also be used and/or particular elements might be implemented in hardware, software (including portable software, such as applets) or both. Further, connection to other computing devices such as network input/output devices may be employed.
[0172] Storage media and computer readable media for containing code, or portions of code, can include any appropriate media known or used in the art, including storage media and communication media, such as but not limited to volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage and/or transmission of information such as computer readable instructions, data structures, program modules or other data, including RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other medium which can be used to store the desired information and which can be accessed by the a system device. Based on the disclosure and teachings provided herein, a person of ordinary skill in the art will appreciate other ways and/or methods to implement the various embodiments.
[0173] The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. It will, however, be evident that various modifications and changes may be made thereunto without departing from the broader spirit and scope of the invention as set forth in the claims.
[0174] Other variations are within the spirit of the present disclosure. Thus, while the disclosed techniques are susceptible to various modifications and alternative constructions, certain illustrated embodiments thereof are shown in the drawings and have been described above in detail. It should be understood, however, that there is no intention to limit the invention to the specific form or forms disclosed, but on the contrary, the intention is to cover all modifications, alternative constructions and equivalents falling within the spirit and scope of the invention, as defined in the appended claims.
[0175] The use of the terms “a” and “an” and “the” and similar referents in the context of describing the disclosed embodiments (especially in the context of the following claims) are to be construed to cover both the singular and the plural, unless otherwise indicated herein or clearly contradicted by context. The terms “comprising,” “having,” “including,” and “containing” are to be construed as open-ended terms (i.e., meaning “including, but not limited to,”) unless otherwise noted. The term “connected” is to be construed as partly or wholly contained within, attached to, or joined together, even if there is something intervening. Recitation of ranges of values herein are merely intended to serve as a shorthand method of referring individually to each separate value falling within the range, unless otherwise indicated herein, and each separate value is incorporated into the specification as if it were individually recited herein. All methods described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The use of any and all examples, or exemplary language (e.g., “such as”) provided herein, is intended merely to better illuminate embodiments of the invention and does not pose a limitation on the scope of the invention unless otherwise claimed. No language in the specification should be construed as indicating any non-claimed element as essential to the practice of the invention.
[0176] Preferred embodiments of this disclosure are described herein, including the best mode known to the inventors for carrying out the invention. Variations of those preferred embodiments may become apparent to those of ordinary skill in the art upon reading the foregoing description. The inventors expect skilled artisans to employ such variations as appropriate, and the inventors intend for the invention to be practiced otherwise than as specifically described herein. Accordingly, this invention includes all modifications and equivalents of the subject matter recited in the claims appended hereto as permitted by applicable law. Moreover, any combination of the above-described elements in all possible variations thereof is encompassed by the invention unless otherwise indicated herein or otherwise clearly contradicted by context.
[0177] All references, including publications, patent applications and patents, cited herein are hereby incorporated by reference to the same extent as if each reference were individually and specifically indicated to be incorporated by reference and were set forth in its entirety herein.
Claims
What is claimed is:
1. A computer-implemented method for storing data, comprising:
under the control of one or more computer systems configured with executable instructions,
selecting a data storage device that is at least:
comprised of data storage media that stores data sequentially; and
capable of responding to a write pointer query by returning a write pointer that indicates a sequential last write position upon the data storage media at which data was most recently written; and
in response to receiving a plurality of write requests, at least:
caching data associated with the write requests in local memory of the computer system;
sequentially recording at least a subset of the cached data upon the data storage media; and
verifying, without flushing or committing the subset of cached data, that the data was written successfully by at least:
querying the data storage device for the write pointer; and
comparing the sequential last write position upon the data storage media indicated by the write pointer with a predetermined sequential destination upon the data storage media for the data being verified.
2. The computer-implemented method of claim 1, wherein the data storage device is configured to use shingled magnetic recording techniques.
3. The computer-implemented method of claim 1, wherein, before selecting the data storage device, adapting the data storage device to be capable of responding to the write pointer query.
4. The computer-implemented method of claim 3, wherein adapting the data storage device to be capable of responding to the write pointer query comprises adapting the data storage device to be capable of responding to the write pointer query by at least updating a firmware of the data storage device.
5. The computer-implemented method of claim 1, wherein the data storage device persists, in a solid state memory of the data storage device, information about the returned write pointer.
6. The computer-implemented method of claim 1, wherein the plurality of write requests is received by a storage system that is at least associated with a plurality of the data storage devices.
7. A computer-implemented method for storing data, comprising:
under the control of one or more computer systems configured with executable instructions,
receiving a plurality of data write requests; and
processing the plurality of data write requests by at least:
queuing the data write requests in local memory of the one or more computer systems;
recording, upon one or more data storage devices that store data sequentially, data associated with at least a subset of the queued data write requests; and
verifying that the data was written successfully by at least:
querying the data storage device upon which the data was written for a write pointer that indicates a final sequentially written location upon the data storage device; and
comparing the write pointer to a known sequential destination for the data being verified.
8. The computer-implemented method of claim 7, wherein at least a subset of the one or more data storage devices record data using shingled magnetic recording techniques.
9. The computer-implemented method of claim 7, wherein the data write requests are received through a programmatic interface exposed to users connected to the computer system.
10. The computer-implemented method of claim 7, wherein the data are verified as successfully written when a result of the comparing the write pointer to the known sequential destination indicates that the known sequential destination sequentially precedes the write pointer on the data storage device.
11. The computer-implemented method of claim 7, wherein the data associated with the queued data write requests is recorded in an analogous order to that of the queued data write requests.
12. The computer-implemented method of claim 7, wherein the data storage device is capable of random reads.
13. A data storage system, comprising:
one or more processors;
data storage media that is capable of storing data in a sequential fashion; and
memory, including instructions executable by the one or more processors to cause the data storage system to at least:
track a terminal write location of the data storage media that reflects a sequentially furthest location to which the data storage media has been written at a given time;
at a time after receiving an external request for the terminal write location of the data storage media, respond with the tracked terminal write location of the data storage media without flushing or committing the data to the data storage media.
14. The data storage system of claim 13, wherein the data storage media is capable of random data reads.
15. The data storage system of claim 13, wherein the data storage media operates using shingled magnetic recording techniques.
16. The data storage system of claim 13, wherein the external request is received from a second data storage system to which at least the data storage media is connected.
17. The data storage system of claim 13, wherein the terminal write location is persisted in the memory.
18. The data storage system of claim 13, wherein the terminal write location is stored upon the data storage media.
19. The data storage system of claim 13, wherein the terminal write location is tracked within the memory and wherein the instructions cause the data storage system to further respond with the terminal write location from the memory.
20. The data storage system of claim 13, wherein the external request is received through a programmatic interface exposed by the data storage system.
21. One or more non-transitory computer-readable storage media having collectively stored thereon executable instructions that, when executed by one or more processors of a computing resource provider's computer system, cause the computer system to at least:
service data storage requests by at least:
pipelining the data storage requests by storing them in a local memory of the computer system; and
processing, in a specified order, each of the data storage requests by committing data associated with each of the data storage requests to a data storage device to which the data are written by appending the data after a sequentially final location to which the device was previously written; and
confirm that a given subset of the data storage requests were serviced by at least:
requesting, from the data storage device, a write location bookmark that indicates a current sequential write position on the data storage device; and
comparing the current sequential write position of the requested write location bookmark that indicates a current sequential write position on the data storage device with a predetermined intended destination for data associated with the subset of data storage requests on the data storage device.
22. The non-transitory computer-readable storage media of claim 21, wherein the subset of data storage requests is confirmed as serviced when a result of the comparing of the current sequential write position with the predetermined intended destination indicates that the predetermined intended destination sequentially precedes the current sequential write position on the data storage device.
23. The non-transitory computer-readable storage media of claim 21, wherein the data storage device is a hard disk drive that records data using shingled magnetic recording techniques.
24. The non-transitory computer-readable storage media of claim 21, wherein the data storage device returns the write location bookmark in response to the request.
25. The non-transitory computer-readable storage media of claim 24, wherein the write location bookmark is returned without a decrease in transaction rate due to flushing or committing the data during the processing of the data storage requests.
26. The non-transitory computer-readable storage media of claim 21, wherein the predetermined intended destination is determined by the computer system during the processing of the data storage requests.
Patent Citations (89)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US5506809(A) | 1996-04-01 | Csoppenszky et al. | Examiner |
| US5586291(A) | 1996-12-01 | Lasker et al. | Examiner |
| US5751997(A) | 1998-05-01 | Kullick et al. | Applicant |
| US5900007(A) | 1999-05-01 | Nunnelley et al. | Applicant |
| US6023710(A) | 2000-02-01 | Steiner et al. | Applicant |
| US6138126(A) | 2000-10-01 | Hitz | Examiner |
| US6578127(B1) | 2003-06-01 | Sinclair | Examiner |
| US6606629(B1) | 2003-08-01 | DeKoning et al. | Applicant |
| US6768863(B2) | 2004-07-01 | Ando et al. | Applicant |
| US6950967(B1) | 2005-09-01 | Brunnett et al. | Examiner |
| US7076604(B1) | 2006-07-01 | Thelin | Examiner |
| US7120737(B1) | 2006-10-01 | Thelin | Examiner |
| US7269733(B1) | 2007-09-01 | O'Toole | Applicant |
| US7577689(B1) | 2009-08-01 | Masinter et al. | Applicant |
| US7774466(B2) | 2010-08-01 | Coates et al. | Applicant |
| US7814078(B1) | 2010-10-01 | Forman et al. | Applicant |
| US7827201(B1) | 2010-11-01 | Gordon et al. | Applicant |
| US8015158(B1) | 2011-09-01 | Mankovsky et al. | Applicant |
| US8019925(B1) | 2011-09-01 | Vogan et al. | Applicant |
| US8130554(B1) | 2012-03-01 | Linnell | Applicant |
| US8156381(B2) | 2012-04-01 | Tamura et al. | Applicant |
| US8291170(B1) | 2012-10-01 | Zhang et al. | Applicant |
| US8352439(B1) | 2013-01-01 | Lee et al. | Applicant |
| US8370315(B1) | 2013-02-01 | Efstathopoulos et al. | Examiner |
| US8464133(B2) | 2013-06-01 | Grube et al. | Applicant |
| US8595596(B2) | 2013-11-01 | Grube et al. | Applicant |
| US8620870(B2) | 2013-12-01 | Dwarampudi et al. | Applicant |
| US8699159(B1) | 2014-04-01 | Malina | Examiner |
| US2002/0055942(A1) | 2002-05-01 | Reynolds | Applicant |
| US2002/0122203(A1) | 2002-09-01 | Matsuda | Examiner |
| US2002/0186844(A1) | 2002-12-01 | Levy et al. | Applicant |
| US2003/0033308(A1) | 2003-02-01 | Patel et al. | Applicant |
| US2004/0003272(A1) | 2004-01-01 | Bantz et al. | Applicant |
| US2004/0098565(A1) | 2004-05-01 | Rohlman et al. | Examiner |
| US2005/0114338(A1) | 2005-05-01 | Borthakur et al. | Applicant |
| US2005/0160427(A1) | 2005-07-01 | Ustaris | Applicant |
| US2005/0203976(A1) | 2005-09-01 | Hyun et al. | Applicant |
| US2005/0267935(A1) | 2005-12-01 | Gandhi et al. | Applicant |
| US2006/0095741(A1) | 2006-05-01 | Asher et al. | Examiner |
| US2006/0272023(A1) | 2006-11-01 | Schmeidler et al. | Applicant |
| US2007/0011472(A1) | 2007-01-01 | Cheng | Applicant |
| US2007/0050479(A1) | 2007-03-01 | Yoneda | Examiner |
| US2007/0079087(A1) | 2007-04-01 | Wang et al. | Applicant |
| US2007/0101095(A1) | 2007-05-01 | Gorobets | Applicant |
| US2007/0156842(A1) | 2007-07-01 | Vermeulen et al. | Applicant |
| US2007/0174362(A1) | 2007-07-01 | Pham et al. | Applicant |
| US2007/0198789(A1) | 2007-08-01 | Clark et al. | Applicant |
| US2008/0059483(A1) | 2008-03-01 | Williams et al. | Applicant |
| US2008/0068899(A1) | 2008-03-01 | Ogihara et al. | Applicant |
| US2008/0120164(A1) | 2008-05-01 | Hassler | Applicant |
| US2008/0177697(A1) | 2008-07-01 | Barsness et al. | Applicant |
| US2008/0294764(A1) | 2008-11-01 | Wakako | Applicant |
| US2009/0083476(A1) | 2009-03-01 | Pua et al. | Applicant |
| US2009/0157700(A1) | 2009-06-01 | Van Vugt | Applicant |
| US2009/0164506(A1) | 2009-06-01 | Barley et al. | Applicant |
| US2009/0213487(A1) | 2009-08-01 | Luan et al. | Examiner |
| US2009/0234883(A1) | 2009-09-01 | Hurst et al. | Applicant |
| US2009/0254572(A1) | 2009-10-01 | Redlich et al. | Applicant |
| US2009/0265568(A1) | 2009-10-01 | Jackson | Applicant |
| US2010/0037056(A1) | 2010-02-01 | Follis et al. | Applicant |
| US2010/0094819(A1) | 2010-04-01 | Bornhoevd et al. | Applicant |
| US2010/0169544(A1) | 2010-07-01 | Eom et al. | Applicant |
| US2010/0217927(A1) | 2010-08-01 | Song et al. | Applicant |
| US2010/0235409(A1) | 2010-09-01 | Roy et al. | Applicant |
| US2010/0242096(A1) | 2010-09-01 | Varadharajan et al. | Applicant |
| US2011/0026942(A1) | 2011-02-01 | Naito | Examiner |
| US2011/0035757(A1) | 2011-02-01 | Comer | Applicant |
| US2011/0058277(A1) | 2011-03-01 | de la Fuente et al. | Applicant |
| US2011/0078407(A1) | 2011-03-01 | Lewis | Applicant |
| US2011/0225417(A1) | 2011-09-01 | Maharajh et al. | Applicant |
| US2011/0247074(A1) | 2011-10-01 | Manring et al. | Applicant |
| US2011/0276656(A1) | 2011-11-01 | Knapp et al. | Applicant |
| US2011/0282839(A1) | 2011-11-01 | Paksoy et al. | Applicant |
| US2011/0289383(A1) | 2011-11-01 | Dhuse et al. | Applicant |
| US2012/0079562(A1) | 2012-03-01 | Anttila et al. | Applicant |
| US2012/0137062(A1) | 2012-05-01 | Arges et al. | Examiner |
| US2012/0143830(A1) | 2012-06-01 | Cormode et al. | Applicant |
| US2012/0166576(A1) | 2012-06-01 | Orsini et al. | Applicant |
| US2012/0173392(A1) | 2012-07-01 | Kirby et al. | Applicant |
| US2012/0210092(A1) | 2012-08-01 | Feldman | Examiner |
| US2012/0233432(A1) | 2012-09-01 | Feldman et al. | Applicant |
| US2012/0306912(A1) | 2012-12-01 | Blanco | Examiner |
| US2012/0311260(A1) | 2012-12-01 | Yamagiwa et al. | Applicant |
| US2013/0046974(A1) | 2013-02-01 | Kamara et al. | Applicant |
| US2013/0254166(A1) | 2013-09-01 | Kottomtharayil | Applicant |
| US2013/0290263(A1) | 2013-10-01 | Beaverson et al. | Applicant |
| US2014/0068208(A1) | 2014-03-01 | Feldman | Examiner |
| US2014/0149794(A1) | 2014-05-01 | Shetty et al. | Applicant |
| US2014/0161123(A1) | 2014-06-01 | Starks et al. | Applicant |
Non-Patent Literature (18)
- Design Issues for a Shingled Write Disk Sysem by Amer, Long, Miller, Paris, and Schwartz; Published 2010 by IEEE pp. 1, 3, 4, and 7.Examiner
- Cisco Standalone HDD Firmware Update Version 3.0—IBM Servers; Published Nov. 16, 2010, p. 1.Examiner
- The Authoritative Dictionary of IEEE Standards Terms; IEEE; Seventh Edition; 2000 p. 836.Examiner
- The Design and Implementation of a Log-Structured File System; Rosenblum and Osterhout; UC Berkley; 1992.Examiner
- Merriam Webster Dictionary Online; Current Edition; www.merriam-webster.com/dictionary.Examiner
- Firmware Updates for Seagate Products; Author Unknown; As published Feb. 2012 at http://knowledge.seagate.com/articles/en—US/FAQ/207931en.Examiner
- How to Leverage an API for Conferencing by Roos; As published Jan. 2012 at http://money.howstuffworks.com/business-communications/how-to-leverage-an-api-forG-coonferencing1.htm.Examiner
- Memory Systems, Cache DRAM, Disk; Bruce Jacobs; Morgan Kaufman 2008.Examiner
- International Search Report and Written Opinion mailed Feb. 14, 2014, in International Patent Application No. PCT/US2013/053828, filed Aug. 6, 2013.Applicant
- International Search Report and Written Opinion mailed Mar. 6, 2014, in International Patent Application No. PCT/US2013/53852, filed Aug. 6, 2013.Applicant
- International Search Report and Written Opinion mailed Feb. 14, 2014, in International Patent Application No. PCT/US2013/053853, filed Aug. 6, 2013.Applicant
- Anonymous, “Hash Tree,” from Wikipedia.org at http://en.wikipedia.org/wiki/Hash—tree (1 page), retrieved Jul. 12, 2012.Applicant
- “Checksum,” from Wayback/Wikipedia at en.wikipedia.org/wiki/checksum (5 pages), retrieved Mar. 2011.Applicant
- “Error Correction,” from Wayback/Wikipedia.org at en.wikipedia.org/wiki/Error-correcting—code (7 pages), retrieved Sep. 2010.Applicant
- Gibson et al., “Directions for Shingled-Write and Two-Dimensional Magnetic Recording System Architectures: Synergies with Solid-State Disks (CMU-PDL-09-104),” Carnegie Mellon University Research Showcase, Parallel Data Laboratory, Research Centers and Institutes:1-3, May 1, 2009.Applicant
- Micheloni et al., “Inside NAND Flash Memories,” Springer First Edition (ISBN 978-90-481-9430-8):40-42, Aug. 2010.Applicant
- Rosenblum et al., “The Design and Implementation of a Log-Structured File System,” University of California at Berkley, ACM Transactions on Computer Systems Volume/Issue 10(1):26-52, Feb. 1992.Applicant
- Yu et al., “Exploiting sequential access when declustering data over disks and MEMS-based storage,” Distributed and Parallel Databases, Springer Volume/Issue 19(2-3):147-168, May 2006.Applicant