US 5,774,682 AGrant
System for Concurrent Cache Data Access by Maintaining and Selectively Merging Multiple Ranked Part Copies
Issue Date:1998-06-30
•18 Claims
•5 Drawing Sheets
Abstract
A system for providing multiple hosts with concurrent access to cached data by selectively generating, maintaining, modifying, and consolidating multiple versions of data items in cache memory to efficiently accommodate data access requests by the hosts. Data associated with a logical track is represented in cache by a number of cache track image parts. Each part represents one or more records in cache, where multiple parts may exist in cache for the same logical track. The provision of multiple parts supports concurrent access by multiple operations or "processes" to data associated with a track. Namely, each part is given a "status" selected from a predetermined catalog of statuses; the assigned status thus establishes the permissible manner of accessing that part. Depending upon a part's status, the part may be used by one process (e.g. Read) or by multiple processes concurrently (e.g. Read and Destage). Other part statuses dedicate a part to a single process (e.g. Write). Multiple parts, not being actively used by a process, may be combined by a merge function to represent the most current image of the track.
Metadata
Assignee
- International Business Machines Corporation
Inventors
- Michael Thomas Benhase
- David Alan Burton
- Marshall Heyman
- John Norbert McCauley
- Robert Louis Morton
Application Information
Application Number:US 5702801
Filing Date:1995-12-11
Priority Date:1995-12-11
Art Unit:273
Classifications
IPC:
G06F 1202G06F 1314
Field of Search:
395364711308;449;440;465;463;853;182.03;200.49;309;500;381DIG. 1;DIG. 2113;122
Patent Drawings (5 sheets)
Description
Background of the Invention
1. Field of the Invention
The present invention relates to the access of data in a storage subsystem by multiple hosts. More particularly, the invention concerns a system to provide multiple hosts with concurrent access to cached data by selectively generating, maintaining, modifying, and consolidating multiple versions of data items in cache memory to efficiently accommodate data access requests by the hosts.
2. Description of the Related Art
In storage subsystems that implement RAID architecture, disk resident data is protected from loss by a form of data redundancy known as "parity." This redundancy ensures that if a storage device fails, the data can still be reconstructed. Data sectors and parity sectors are distributed across the devices in the array. The data sectors and their corresponding parity sectors are grouped in such a way as to afford continuous availability and maximized update performance. New parity is calculated whenever a sector containing modified data is destaged.
To maintain consistency within the parity domain, the parity sectors must be updated at the same time as their respective data sectors; it is essential that the update of the data and parity sectors be atomic to ensure data integrity. This is accomplished by using a locking mechanism. The unit of data under control of the lock is known as a "segment." Depending upon the granularity of the lock, data availability and write access may be affected while a destage is in progress.
Generally, a storage subsystem's throughput is constrained unless that system inherently supports continuous data availability despite concurrent stage/destage operations and simultaneous access by multiple hosts. One problem in locking systems concerns the updating of data undergoing a destage operation. In particular, when modified data is being destaged, a lock may be applied to each segment (e.g. logical track) of the data. The subsystem will not accept a subsequent write to the segment until the lock is released, i.e. when the destage completes. The desired write operation is deferred by "control unit initiated command retry"; the command chain remains pending at the subchannel until the subsystem presents ending status. The busy subchannel will prevent any further operations for this logical device from being initiated by this host until the command chain is completed. As a result, data being destaged from cache may be accessible to host Read operations, but it cannot be updated for the duration of the destage due to the lock placed upon the data segment.
Another problem in known systems may occur with a partial-Read-miss. A Read-miss occurs when the cache cannot satisfy the host request. If a Read-miss occurs for a track that has modified data in cache, the modified data will be first destaged; when the destage completes, the track will be staged to cache to satisfy the Read request. The command chain remains pending at the subchannel until the data is available in cache. As a result, a partial-Read-miss may force destaging of modified data before the Read request can be satisfied, resulting in significant delays to host Read and Write operations.
The segment locking approach to cache management can be especially problematic with log structured array ("LSA") storage. In particular, destaging with LSA storage is typically not performed until a complete data segment is accumulated. Thus, when destaging does finally occur, it takes longer to complete. Therefore, other processes are put at a disadvantage to the destaging process, because the other processes cannot accurately anticipate the locking of the destaged data segment, and because the lock is maintained for longer to the exclusion of the other processes.
Summary of the Invention
Broadly, the present invention concerns a system for providing multiple hosts with concurrent access to cached data by selectively generating, maintaining, modifying, and consolidating multiple versions of data items in cache memory to efficiently accommodate data access requests by the hosts. The cached data may originate from nonvolatile storage, such as an array of magnetic data storage disks. In this environment, data associated with logical tracks of the disks are represented in cache by a number of track parts ("parts"). Each part in cache memory represents one or more records of a track stored on disk. A part may represent some or all data fields of the represented records. Multiple parts existing in cache memory may include portions of the same logical track; however, it is possible for a track to be only partially represented in cache.
The facility to concurrently support multiple parts enables multiple operations or "processes" to simultaneously access data associated with a track. Namely, each part is given a "status" selected from a predetermined catalog of recognized statuses; the assigned status thus establishes the permissible manner of accessing that part. Depending upon a part's status, the part may be used by one process (e.g. Read) or by multiple processes concurrently (e.g. Read and Destage). Other part statuses dedicate a part to a single process (e.g. Write).
At various times, the present invention manages part generation, attends to part maintenance, oversees part modification, and consolidates multiple parts corresponding to the same sector and track. For example, multiple "non-busy" parts may be combined by a merge function to represent the most current image of the track. A "non-busy" part is one that is not being actively used by any processes, where the number of processes actively using the part is indicated by a "number-of-users" attribute of the part.
The invention affords its users with a number of distinct advantages. Chiefly, the invention provides an efficient method for ensuring consistency of cached data. Moreover the invention improves data throughput and response time by enabling multiple users to simultaneously access data. For example, the invention provides consistently high performance Fast Write response because destage operations have no effect on host Read or Write operations, since device updates are performed transparently with respect to the host. Moreover, Write performance is improved significantly, especially over other implementations of RAID-5 architecture.
Brief Description of the Drawings
The nature, objects, and advantages of the invention will become more apparent to those skilled in the art after considering the following detailed description in connection with the accompanying drawings, in which like reference numerals designate like parts throughout, wherein:
FIG. 1 is a block diagram of the hardware components and interconnections of the invention;
FIG. 2 is a perspective view of an exemplary data storage medium in accordance with the invention;
FIG. 3 is a flowchart of a write routine pursuant to the invention;
FIG. 4 is a flowchart of a destage routine pursuant to the invention;
FIG. 5 is a flowchart of a stage routine pursuant to the invention; and
FIG. 6 is a flowchart of a read routine pursuant to the invention.
Detailed Description of the Preferred Embodiments
Structure
FIG. 1 illustrates an exemplary hardware environment for implementing the invention. Although other arrangements may be used, the illustrated example concerns a memory system 100 including multiple hosts 101-103, a controller 104, a cache memory 108, a cache directory 106, and a memory subsystem 110.
The memory subsystem 110 comprises a non-volatile storage subsystem, accessible to the hosts 101-103 by the controller 104 and cache memory 108. In a preferred embodiment, the memory subsystem 110 comprises a group of magnetic digital data storage disks configured according to RAID-5 architecture.
The subsystem 110 is coupled to the cache memory 108, which preferably comprises a fast-access volatile memory such as RAM. The cache memory 108 is coupled to a controller 104, which maintains the cache memory 108. Preferably, the controller 104 comprises a RISC processor such as the INTEL brand i 960.TM., or another suitable processing unit. As an alternative, some or all of the functions of the controller 104 may be subsumed by operations of the hosts 101-103. The controller 104 may include or have access to additional data storage (not shown) of its own, for the purpose of storing a sequence of machine-readable programming instructions executed by the controller 104 during operation of the invention as described below. In this respect, such data storage may be provided by a RAM module accessible to the controller, or by a data storage medium such as a computer diskette 200 (FIG. 2). Alternatively, the instructions may also be stored on a DASD array, magnetic tape, conventional "hard drive", electronic read-only memory, optical storage device, paper "punch" cards, or another data storage device. In an illustrative embodiment of the invention, the computer-executable instructions may comprise lines of compiled C++language code.
In addition to the cache memory 108, the controller 104 is also coupled to the cache directory 106 and the hosts 101-103. Each of the hosts 101-103 comprises a "user" of the memory subsystem 110, such as a mainframe computer, personal computer, workstation, user interface device, memory storage device, another cache process, or another suitable system user.
Operation
Operation of the system 100 is performed according to a process aspect of the invention, described as follows.
Parts & Part Status
As mentioned above, the invention contemplates multiple track parts ("parts"), each part providing a cache representation of one or more records of a track stored on disk. A part may represent some or all data fields of the records it represents. Multiple parts existing in cache may include parts of the same logical track; however, it is possible for a track to be only partially represented in cache. One advantage of the invention is that the cache may contain multiple parts containing portions of the same logical track.
The provision of multiple parts supports concurrent access to data associated with a track by multiple operations or "processes". Namely, each part is assigned one status from a predetermined catalog of recognized statuses, thus establishing the permissible manner of accessing that part. Depending upon a part's status, the part may be used by one process (e.g. Read) or by multiple processes concurrently (e.g. Read and Destage). Other part statuses dedicate a part to a single process (e.g. Write).
Following is an exemplary catalog of part statuses, illustrating one embodiment of the invention.
A. Primary
This status denotes a part that has been staged into the cache 108, but has not been modified while cached. Alternatively, the Primary part may contain modified data that has been marked as unmodified following destage completion. The Primary part may be used to satisfy host Read requests, for example.
B. Stage
This status identifies a part that is in the process of being staged into the cache 108.
C. Primary-Write
This status concerns a part that is in the process of being modified, i.e. being written to by one of the hosts 101-103 via the controller 104.
D. Primary-Modified
This status denotes a part that has been modified after it was staged to cache. During the process of writing to a part, the part is Primary-Write; following the writing operation, the part assumes Primary-Modified status.
E. Primary-Modified-Merge-Pending
This status describes a Primary-Modified part that has been slated for merging with another part. Merging operations are discussed in greater detail below.
F. Destage-Pending
This status attaches to a modified part that has been slated for future destaging, although the destaging operation has not yet begun.
G. Primary-In-Destage
This status refers to a modified part that is undergoing the process of being destaged to the memory subsystem 110. As discussed above, after a Primary part has been modified, it becomes a Primary-Modified part; then, before the process of destaging this part, the part becomes Destage-Pending. Subsequently, during the destage operation, the part assumes Primary-In-Destage status.
H. Previous
This status concerns a part that is maintained in the cache 108 for the purpose of supporting parity operations. After a part is modified in the cache 108 it becomes a Primary-Modified part. The original (unmodified) version of that part is maintained in the cache 108 as a Previous status part. This part is used in non-failure situations to allow parity to be updated in accordance with new data without having to read the old data from the memory subsystem 110, thereby improving performance. In particular, the new parity is computed as shown in Equation 1, below:
where:
"Old Parity" represents the existing parity computation, to be updated by the "New Parity";
"New Data" represents the new part, an update of an old part; and
"Old data" represents the old part which is being updated by the New Data, and is represented in cache memory by a Previous status part. Therefore,, whenever a cached part is updated, the part is retained in cache as a Previous status part to avoid the need to access the memory subsystem 110 when computing new parity for the updated part.
I. Previous-In-Destage
This status, like the Previous status, concerns a part that is maintained in the cache 108 for the purpose of supporting parity operations. When a modified part is undergoing destaging to the cache 108 it becomes a Primary-In-Destage part. The original (unmodified) version of that part is maintained in the cache 108 as a Previous-In-Destage status part during this time. Before the process of destaging the part, the part is merely Previous status; during the destage operation, the Previous part assumes Previous-In-Destage status.
As described above, the present invention provides a catalog of allowable part statuses used in maintaining cache consistency. The nature and significance of these part statuses will become more apparent with the following description of the procedures for managing and consolidating parts.
Part Attributes
In addition to a status as listed above, each cached part may have certain attributes. For example, each part may include a "number-of-users" attribute, indicating the number of different users (e.g. the hosts 101-103) that are currently accessing the part. If a part's number-of-users attribute is non-zero, the part is considered to be "busy. "
Write
FIG. 3 depicts a sequence 300 of tasks for performing Write operations in a hardware environment such as the system 100, using the catalog of part statuses described above. In this context, "Write" operations involve making modifications to cached data. Thus, the sequence 300 is performed whenever a user (e.g. one of the hosts 100-103) initiates an attempt to modify data contained in the cache 108.
The user begins the routine 300 in task 302 by issuing a Write request accompanied by an identification of one or more data parts to be written. Then, in task 304, the controller 104 begins to write the data parts to the cache 108. One advantage of the invention is that host Write operations are not inhibited at any time; namely, data is immediately written to the cache 108, with consolidation by merging being performed later if possible.
In response to task 304, the controller 104 in task 305 places the parts-to-be-written in Primary-Write status. Then, the controller 104 in query 306 asks whether the operation of writing the data parts to the cache 108 is complete. If not, the writing continues in task 307.
After an affirmative answer to query 306, performance of the merge rules is permitted, as shown by task 310. Parts being considered for possible merger are referred to as "candidate" parts. For each Primary-Write candidate data part, the controller 104 in query 312 asks whether there are any "partners". A "partner" refers to another data part that represents one or more records from the subsystem 110 that are also represented by the candidate data part. For example, the partner and candidate may concern a completely identical parcel of data, or the partner and candidate parts may represent records in common while also representing non-overlapping data records. Searching for partner parts is performed because there is a possibility that the candidate and partner may be merged to operate the cache 108 more efficiently.
If no partner for the candidate part is found, the candidate data part is changed from Primary-Write to Primary-Modified status in task 314. Following task 314, the routine 300 ends in task 324.
If query 312 finds a partner for the candidate, however, it may be possible to consolidate parts by merging the candidate with the partner. Therefore, the controller 104 in task 316 applies certain merge rules, discussed in greater detail below. The result of applying these merge rules may be creation of a Primary-Modified-Merge-Pending part (task 322), a Primary-Modified part (task 320), or a Primary-In-Destage part (task 318). After applying the merge rules, the routine 300 ends in task 324.
Write Merge Rules
As stated above, the Write operation merge rules seek to combine a candidate Primary-Write part with a partner of another status. To perform a merge operation, the candidate and partner must both be non-busy, i.e. having a number-of-users attribute equal to zero. If one or both parts are busy, the process may wait until both parts are non-busy.
The manner of merging the two parts depends upon the status of the partner, as shown below.
1. Primary-Modified
If the partner is a Primary-Modified part, the candidate and partner are merged, and the resultant part is given Primary-Modified status.
2. Primary-Modified-Merge-Pending
If the partner is a Primary-Modified-Merge-Pending part, no merging action is taken immediately. However, the candidate is given Primary-Modified-Merge-Pending status pending a subsequent merge with the partner part. Later, after the first merge operation completes, the candidate is merged with the newly merged part to form a consolidated Primary-Modified part.
3. Primary-In-Destage
If the partner is a Primary-In-Destage part, no merging action is taken. The candidate part's status is changed to Primary-Modified. Potential merging, if any, between the candidate and partner parts will be accomplished next time the merge rules are applied.
4. Destage-Pending
If the partner is a Destage-Pending part, the candidate and partner are merged to yield a consolidated part having Primary-In-Destage status.
5. Primary
If the partner is a Primary part, the candidate and partner are merged to yield a consolidated part having Primary-Modified status. A copy of the partner is made in the cache 108 and given the Previous status to preserve the ability to perform parity reconstruction of the corresponding data.
6. Stage
If the partner is a Stage part, no merging action is taken immediately. The candidate part, however, is slated for merging with the partner after its staging is complete, whereupon the combined part will be given Primary-Modified status. In the mean time, the candidate part assumes Primary-Modified-Merge-Pending status.
7. Previous-In-Destage
If the partner is a Previous-In-Destage part, no merging action is taken. The candidate part's status is changed to Primary-Modified.
8. Previous
If the partner is a Previous part, no merging action is taken. The candidate part's status is changed to Primary-Modified.
9. Primary-Write
For a Primary-Write candidate, the status is not valid for potential partners, i.e. it cannot exist. Only one Primary-write part can exist at one time.
Destage
FIG. 4 depicts a sequence 400 of tasks for performing Destage operations in a hardware environment such as the system 100, using the catalog of part statuses described above. Generally, "Destage" operations concern data stored in the cache 108 that has been modified after it was originally copied (i.e. "staged") into the cache 108 from the memory subsystem 110. More particularly, destage operations involve writing this modified data back to the subsystem 110 to update the subsystem 110. Generally, unmodified records are not destaged unless they are needed to connect two more disjoint parts containing modified data.
The user begins the routine 400 in task 402 by issuing a Destage request accompanied by an identification of one or more data parts to be destaged. The controller 104 receives the user's request in task 404. The controller 104 in query 406 asks whether all identified parts are present in the cache 108. If not, the identified parts present in the cache 108 are given Destage-Pending status in task 408, and the missing parts are staged in task 410. The Stage routine 500 (FIG. 5) is described in greater detail below.
After all parts identified in task 404 are present, being either initially present or present after staging in task 410, the controller 104 in task 412 begins destaging the parts to the subsystem 110. Accordingly, in task 414 the controller 104 assigns these parts Primary-In-Destage status.
In query 416, the controller 104 asks whether the destage operation is complete. If not, destaging continues in task 418. Otherwise, when all data parts have been successfully written to the subsystem 110, performance of the merge rules is permitted, as shown by task 422. First, the controller 104 in query 424 asks whether, for each Primary-In-Destage "candidate" data part, there is a partner part in cache 108. As mentioned above, searching for partner parts is performed because there is a possibility that the candidate and partner may be merged to operate the cache 108 more efficiently.
If no partner for the candidate part is found, the candidate data part is made a Primary status part, as shown by task 426. Following task 426, the routine 400 ends in task 432. If query 424 finds a partner for the candidate, however, it may be possible to consolidate parts by merging the candidate with task 428 applies certain merge controller 104 in task 428 applies certain merge rules, discussed in greater detail below. The result of applying these merge rules may be creation of a Primary part (task 426) or a Primary-Modified part (task 430). After applying the merge rules, the routine 400 ends in task 432.
Destage Merge Rules
As stated above, the Destage operation merge rules seek to combine a candidate Primary-In-Destage part with a partner of another status. To perform a merge operation, the candidate and partner must both be non-busy, i.e. having a number-of-users attribute equal to zero. If one or both parts are busy, the process may wait until both parts are non-busy.
The manner of merging the two parts depends upon the status of the partner, as shown below.
1. Primary-Write
If the partner is a Primary-Write part, no merging action is taken. The Primary-In-Destage candidate is switched to Primary status.
2. Primary-Modified
If the partner is a Primary-Modified part, the candidate and partner are merged, and the resultant part is given Primary-Modified status. Any displaced data is used to generate a part in the cache 108 with Previous status, to preserve the ability to assist in parity construction of the corresponding data.
3. Primary-Modified-Merge-Pending
If the partner is a Primary-Modified-Merge-Pending part, no merging action is taken. The Primary-In-Destage candidate is switched to Primary status.
4. Primary-In-Destage
For a Primary-In-Destage candidate, this status is not valid for potential partners, i.e. it cannot exist. Only one Primary-In-Destage part can exist at one time.
5. Destage-Pending
For a Primary-In-Destage candidate, this status is not valid, i.e. cannot exist.
6. Primary
If the partner is a Primary part, the candidate and partner are merged to yield a consolidated part having Primary status. Any overlapping data is discarded. A copy of the partner is made in the cache 108 and given the Previous status to preserve the ability to perform parity reconstruction of the corresponding data.
7. Stage
For a Primary-In-Destage candidate, this is not a valid status for potential partners.
8. Previous-In-Destage
If the partner is a Previous-In-Destage part, no merging action is taken, but the partner is deleted.
9. Previous
If the partner is a Previous part, no merging action is taken. The candidate part's status is changed to Primary.
Stage
FIG. 5 depicts a sequence 500 of tasks for performing Stage operations in a hardware environment such as the system 100, using the catalog of part statuses described above. Generally, "Stage" operations are performed to copy data from the subsystem 110 into the cache 108, such as in response to a Read-miss.
The Stage routine 500 is initiated in task 502 in response to a Read-miss or another request for data from the subsystem 110 that is not present in the cache 108. This data consists of one or more identified data parts. The controller 104 in task 502 begins the Stage procedure, and in task 506 designates the identified data parts as Stage status parts.
In query 508, the controller 104 asks whether the Stage operation is complete. If not, staging continues in task 510. Otherwise, when all data parts have been successfully copied to cache 108 from the subsystem 110, performance of the merge rules is permitted as shown by task 514.
First, the controller 104 in query 516 asks whether, for each Primary "candidate" data part, there is a partner part in cache 108. As mentioned above, searching for partner parts is performed because there is a possibility that the candidate and partner may be merged to operate the cache 108 more efficiently.
If no partner for the candidate part is found, the candidate data part becomes a Primary status part, as shown by task 518. Following task 51 8, the routine 500 ends in task 526.
If query 516 finds a partner for the candidate, however, it may be possible to consolidate parts by merging the candidate with the partner. Therefore, the controller 104 in task 520 applies certain merge rules, discussed in greater detail below. The result of applying these merge rules may be creation of a Primary part (task 518), a Primary-In-Destage part (task 522), or a Primary-Modified part (task 524). After applying the merge rules, the routine 500 ends in task 526.
Stage Merge Rules
As stated above, the Stage operation merge rules seek to combine a candidate Stage part with a partner of another status. To perform a merge operation, the candidate and partner must both be non-busy, i.e. having a number-of-users attribute equal to zero. If one or both parts are busy, the process may wait until both parts are non-busy.
The manner of merging the two parts depends upon the status of the partner, as shown below.
1. Primary-Write
If the partner is a Primary-Write part, no merging action is taken since the Primary-Write part is undergoing a write operation. However, the candidate is given Primary status.
2. Primary-Modified
If the partner is a Primary-Modified part, the candidate and partner are merged, and the resultant part is given Primary-Modified status. Any overlapping data is used to generate a Previous status part.
3. Primary-Modified-Merge-Pending
If the partner is a Primary-Modified-Merge-Pending part, the candidate and partner are merged, and the resultant part is given Primary-Modified status. Any overlapping data is used to generate a Previous status part.
4. Primary-In-Destage
If the partner is a Primary-In-Destage part, no merging action is taken. The candidate part's status is changed to Primary.
5. Destage-Pending
If the partner is a Destage-Pending part, the candidate and partner are merged, with the resulting part comprising a Primary-In-Destage part. 6. Primary
If the partner is a Primary part, the candidate and partner are merged to yield a consolidated part having Primary status.
7. Stage
For a Stage candidate, this status is not a valid status for potential partners.
8. Previous-In-Destage
If the partner is a Previous-In-Destage part, no merging action is taken. The candidate part's status is changed to Primary.
9. Previous
If the partner is a Previous part, no merging action is taken. The candidate part's status is changed to Primary.
Read
FIG. 6 depicts a sequence 600 of tasks for performing Read operations in a hardware environment such as the system 100, using the catalog of part statuses described above. In this context, "Read" operations involve reading data from the cache 108 or the subsystem 100. Thus, the sequence 600 is performed whenever a user (e.g. one of the hosts 100-103) initiates an attempt to read data from the subsystem 110, some, all, or none of this data may already be stored in the cache 108.
The user begins the routine 600 in task 602 by issuing a Read request accompanied by an identification of one or more data parts to be read. After the controller 104 receives the Read request in task 604, query 606 asks whether all identified parts are present in the cache 108. If so, the controller 104 conducts a normal Read operation, simply reading each requested data part from the cache 108.
If the cache 108 does not already contain all identified parts, however, the controller 104 initiates tasks 610 and 612 in parallel. In task 610, the controller 104 stages the missing parts; this may be achieved using the Stage routine 500 described above. Reading of these parts constitutes "Read-misses", because the parts are not already contained in the cache 108. When they are staged to cache 108, the staged parts are made available to the user.
Concurrently with task 610, the controller 104 in task 612 provides the user with access to the available parts, i.e. the parts already contained in the cache 108. Accordingly, reading of these parts constitutes "Read-hits". For each requested part, the controller 104 supplies the user with the "highest-ranked" part contained in cache 108, irrespective of the parts' number-of-users attribute. The ranking of a given part relative to other parts is based upon the parts' statuses, as ranked in Table 1, below.
Thus, for example, if Primary-Modified, Destage-Pending, and Primary versions of a part exist in the cache 108, the controller 104 in task 612 provides a Read-hit output of the Primary-Modified part.
In query 616, the controller 104 asks whether all parts have been effectively read. If not, the controller in query 614 asks whether the missing parts are available, i.e. have been staged yet. If not, task 615 waits for a predetermined length of time. When query 614 finds that the missing parts are available, task 612 provides the user with access to these parts. The staged parts are given Primary status, as shown by task 611.
When query 616 determines that all identified parts have been read, performance of the merge rules is permitted, as shown by task 618. First, for each "candidate" data part, the controller 104 in query 620 asks whether there are any partners, where a "partner" refers to another data part that pertains to some of the same data from the subsystem 110 as the candidate data part. If no partner for the candidate part is found, the status of the candidate data part is not changed, as shown by the advancement of query 620 to task 630.
If query 620 finds a partner for the candidate, however, it may be possible to consolidate parts by merging the candidate with the partner. Therefore, the controller 104 in task 622 applies certain merge rules, discussed in greater detail below. The result of applying these merge rules may be creation of a Primary part (task 624), a Primary-Modified part (task 626), or a Primary-In-Destage part (task 628). After applying the merge rules, the routine 600 ends in task 630.
Read Merge Rules
As stated above, the Read operation merge rules seek to combine a candidate part with a partner of the same or another status. To perform a merge operation, the candidate and partner must both be non-busy, i.e. having a number-of-users attribute equal to zero. If one or both parts are busy, the process may wait until both parts are non-busy.
The candidate part may have any status possible, i.e. Primary-Write, Primary-Modified, Primary-Modified-Merge-Pending, Primary-In-Destage, Destage-Pending Primary, Stage, Previous, or Previous-In-Destage. Thus, the manner of merging the candidate and partner depends upon the status of both parts, as shown below.
1. Primary-Write Candidate With Read operations, a Primary-Write candidate cannot exist.
2. Primary-Modified Candidate
A. Primary-Modified Partner
If the partner and candidate are Primary-Modified parts, the candidate and partner are merged, and the resultant part is given Primary-Modified status.
B. Primary-Modified-Merge-Pending Partner
If the candidate is Primary-Modified while the partner is a Primary-Modified-Merge-Pending part, no merging action is taken immediately. However, the candidate is given Primary-Modified-Merge-Pending status pending a subsequent merge with the partner part. Later, after the first merge operation completes, the candidate is merged with the newly merged partner part to form a consolidated Primary-Modified part.
C. Primary-In-Destage Partner
If the candidate is Primary-Modified while the partner is a Primary-In-Destage part, no merging action is taken. The candidate part's status remains Primary-Modified. Potential merging, if any, between the candidate and partner parts will be accomplished next time the merge rules are applied.
D. Destage-Pending Partner
If the candidate is Primary-Modified while the partner is a Destage-Pending part, no merging action is taken. The candidate part's status remains Primary-Modified. Potential merging, if any, between the candidate and partner parts will be accomplished next time the merge rules are applied.
E. Primary Partner
If the candidate is Primary-Modified while the partner is a Primary part, the candidate and partner are merged to yield a consolidated part having Primary-Modified status. A copy of the partner is made in the cache 108 and given the Previous status to preserve the ability to perform parity reconstruction of the corresponding data.
F. Stage Partner
If the candidate is Primary-Modified while the partner is a Stage part, no merging action is taken immediately. The candidate part, however, is slated for merging with the partner; after the partner's staging is complete, the combined part will be given Primary-Modified status. In the mean time, the candidate part assumes Primary-Modified-Merge-Pending status.
G. Previous-In-Destage Partner
If the candidate is Primary-Modified while the partner is a Previous-In-Destage part, no merging action is taken. The candidate part's status is changed to Primary-Modified.
H. Previous Partner
If the candidate is Primary-Modified while the partner is a Previous part, no merging action is taken. The candidate part's status is changed to Primary-Modified.
I. Primary-Write Partner
No merging action is taken here. The candidate part retains Primary-Modifier status.
3. Primary-Modified-Merge-Pending Candidate
With Read operations, a Primary-Modified-Merge-Pending candidate cannot exist. Therefore, this combination is invalid.
4. Primary-In-Destage Candidate
In this case, no merging action is taken. The candidate retains Primary-In-Destage status.
5. Destage-Pending Candidate
With Read operations, a Destage-Pending candidate cannot exist. Therefore, this combination is invalid.
6. Primary Candidate
A. Primary-Write Partner
In this case, no merging action is taken. The candidate retains Primary status.
B. Primary-Modified Partner
Here, the candidate and partner are merged, with the resulting consolidated part assuming Primary-Modified. Any overlapping data becomes a Previous status part.
C. Primary-Modified-Merge-Pending Partner
No merging action is taken here. The candidate part retains Primary status.
D. Primary-In-Destage Partner
No merging action is taken here. The candidate part retains Primary status.
E. Destage-Pending Partner
No merging action is taken here. The candidate part retains Primary status.
F. Primary Partner
Here, the candidate and partner are merged, with the resulting consolidated part assuming Primary status. Any overlapping data is discarded.
G. Stage Partner
No merging action is taken here. The candidate part retains Primary status.
H. Previous-In-Destage Partner
No merging action is taken here. The candidate part retains Primary status.
I. Previous Partner
No merging action is taken here. The candidate part retains Primary status.
7. Stage Candidate
In this case, no merging action is taken. The candidate retains Stage status until its Stage operation completes, as discussed above.
8. Previous Candidate
With Read operations, a Previous candidate cannot exist.
9. Previous-In-Destage Candidate
With Read operations, a Previous-In-Destage candidate cannot exist.
Other Embodiments
While there have been shown what are presently considered to be preferred embodiments of the invention, it will be apparent to those skilled in the art that various changes and modifications can be made herein without departing from the scope of the invention as defined by the appended claims.
Claims
What is claimed is:
1. A method of providing multiple processes with concurrent data access to stored data, comprising the steps of: for each one of at least one item of data, maintaining multiple data parts in cache memory corresponding to the item of data in response to different storage operations involving the item of data; maintaining, for each item of data having one or more corresponding data parts in the cache memory, a listing of status of the data parts, the status of each data part designating a permissible manner of access to that data part by requesting processes; receiving a request from one of the processes for access to the data item; consulting the status listing to identify all data parts corresponding to the requested data item; in response to the request, identifying one of the data parts whose permissible manner of access is compatible with the received request for access to the data item; and providing the requesting process with access to the identified data part.
2. The method of claim 1, further comprising the steps of, in response to a predetermined event, merging selected data parts in the cache memory corresponding to a common data item.
3. The method of claim 1, further comprising the steps of, in response to a predetermined condition, deleting selected data parts.
4. The method of claim 1, each status listing ranking the data parts identified therein, the request for access to the data item comprising a Read request, and the identifying step comprising the steps of identifying a highest-ranked one of the data parts.
5. A method of providing multiple processes with concurrent data access to stored data, comprising the steps of: for an item of data represented in cache memory, maintaining multiple data parts corresponding to the item of data in response to different storage operations; maintaining, for each item of data, a listing of ranked status for the corresponding data parts; receiving a request from one of the processes for access to the data item; consulting the listing to identify all data parts corresponding to the requested data item; identifying in response to the request one of the data parts having a predetermined relationship in status to the other data parts; providing the requesting process with access to the identified data part; and in response to a predetermined event, merging selected data parts corresponding to a common data item, the merging comprising: identifying first and second data parts corresponding to a common data item; determining whether the first and second data parts are currently being used by a process; and applying predetermined merge rules to determine whether merging of the first and second data parts is possible, and if so, merging the first and second data parts to form a consolidated third data part.
6. A method of providing multiple processes with concurrent data access to stored data, comprising the steps of: for an item of data represented in cache memory, maintaining multiple data parts corresponding to the item of data in response to different storage operations; maintaining, for each item of data, a listing of ranked status for the corresponding data parts; receiving a request from one of the processes for access to the data item; consulting the listing to identify all data parts corresponding to the requested data item; identifying in response to the request one of the data parts having a predetermined relationship in status to the other data parts; providing the requesting process with access to the identified data part; and in response to a predetermined event, merging selected data parts corresponding to a common data item, the predetermined event comprising completion of a storage operation involving a data part corresponding to the common item of data.
7. A data storage medium tangibly embodying a program of machine instructions executable by a digital data processing apparatus to perform method steps for providing multiple processes with concurrent data access to stored data, said method steps comprising: for each one of at least one item of data, maintaining multiple data parts in cache memory corresponding to the item of data in response to different storage operations involving the item of data; maintaining, for each item of data having one or more corresponding data parts in the cache memory, a listing of status of the data parts, the status of each data part designating a permissible manner of access to that data part by requesting processes; receiving a request from one of the processes for access to the data item; consulting the status listing to identify all data parts corresponding to the requested data item; in response to the request, identifying one of the data parts whose permissible manner of access is compatible with the received request for access to the data item; and providing the requesting process with access to the identified data part.
8. The data storage medium of claim 7, further comprising the steps of, in response to a predetermined event, merging selected data parts in the cache memory corresponding to a common data item.
9. The data storage medium of claim 7, further comprising the steps of, in response to a predetermined condition, deleting selected data parts.
10. The data storage medium of claim 7, each status listing ranking the data parts identified therein, the request for access to the data item comprising a Read request, and the identifying step comprising the steps of identifying a highest-ranked one of the data parts.
11. A data storage medium tangibly embodying a program of machine instructions executable by a digital data processing apparatus to perform method steps for providing multiple processes with concurrent data access to stored data, said method steps comprising: for an item of data represented in cache memory, maintaining multiple data parts corresponding to the item of data in response to different storage operations; maintaining, for each item of data, a listing of ranked status for the corresponding data parts; receiving a request from one of the processes for access to the data item; consulting the listing to identify all data parts corresponding to the requested data item; identifying in response to the request one of the data parts having a predetermined relationship in status to the other data parts; providing the requesting process with access to the identified data part; and in response to a predetermined event, merging selected data parts corresponding to a common data item, the merging comprising: identifying first and second data parts corresponding to a common data item; determining whether the first and second data parts are currently being used by a process; and applying predetermined merge rules to determine whether merging of the first and second data parts is possible, and if so, merging the first and second data parts to form a consolidated third data part.
12. A data storage medium tangibly embodying a program of machine instructions executable by a digital data processing apparatus to perform method steps for providing multiple processes with concurrent data access to stored data, said method steps comprising: for an item of data represented in cache memory, maintaining multiple data parts corresponding to the item of data in response to different storage operations; maintaining, for each item of data, a listing of ranked status for the corresponding data parts; receiving a request from one of the processes for access to the data item; consulting the listing to identify all data parts corresponding to the requested data item; identifying in response to the request one of the data parts having a predetermined relationship in status to the other data parts; providing the requesting process with access to the identified data part; and in response to a predetermined event, merging selected data parts corresponding to a common data item, the predetermined event comprising completion of a storage operation involving a data part corresponding to the common item of data.
13. A data storage system, comprising: at least one host system performing multiple processes; a cache memory; and a controller, coupled to said at least one host system and said cache memory, said controller being programmed to execute a program of machine instructions to perform method steps for providing the multiple processes with concurrent data access to stored data, said method steps comprising: for each one of at least one item of data, maintaining multiple data parts in cache memory corresponding to the item of data in response to different storage operations involving the item of data; maintaining, for each item of data having one or more corresponding data parts in the cache memory, a listing of status of the data parts, the status of each data part designating a permissible manner of access to that data part by requesting processes; receiving a request from one of the processes for access to the data item; consulting the status listing to identify all data parts corresponding to the requested data item; in response to the request, identifying one of the data parts whose permissible manner of access is compatible with the received request for access to the data item; and providing the requesting process with access to the identified data part.
14. The system of claim 13, further comprising the steps of, in response to a predetermined event, merging selected data parts in the cache memory corresponding to a common data item.
15. The system of claim 14, the merging step comprising the steps of: identifying first and second data parts corresponding to a common data item; determining whether the first and second data parts are currently being used by a process; and applying predetermined merge rules to determine whether merging of the first and second data parts is possible, and if so, merging the first and second data parts to form a consolidated third data part.
16. The system of claim 14, wherein the predetermined event comprises completion of a storage operation involving a data part corresponding to the common item of data.
17. The system of claim 13, further comprising the steps of, in response to a predetermined condition, deleting selected data parts.
18. The system of claim 13, each status listing ranking the data parts identified therein, the request for access to the data item comprising a Read request, and the identifying step comprising the steps of identifying a highest-ranked one of the data parts.