US 5,608,903 AGrant
Method and Apparatus for Moving Subtrees in a Distributed Network Directory
Issue Date:1997-03-04
•21 Claims
•4 Drawing Sheets
Abstract
A method of moving leaf objects and subtrees in computer networks that employ a distributed network directory is disclosed. The method employs the existing directories and an authentication procedure for each server. A first object that is under the physical control of the administrator of one partition of the distributed network directory requests access to a second object that is under the physical control of the administrator of another partition of the distributed network directory. The directory verifies that the access control list of the first object includes the second object. The access control list of the second object is then checked to verify that it includes a reference to the first object as an object that is permitted access to the second object. As a result, access is only granted in response to requests from objects that appear in the access control list of the second object. a method of synchronizing the access control lists based upon an authoritative access control list is also disclosed.
Metadata
Assignee
- Novell, Inc.
Inventors
- Ranjan Prasad
- Dale R. Olds
Application Information
Application Number:US 3574660
Filing Date:1994-12-15
Priority Date:1994-12-15
Art Unit:237
Classifications
IPC:
G06F 1730G06F 1200
Field of Search:
395364600;575;800200
Patent Drawings (4 sheets)
Description
Background
The present invention relates to the management of distributed digital network directories, and particularly to moving an object or subtree of a distributed digital network directory within the directory or to another directory.
Technological advances in microelectronics and digital computing systems have resulted in the proliferation of digital computer networks, enabling the distribution of networking services across a wide range of computers participating in the network and over various communications media. Advances in distributing applications have also resulted in a client-server architecture for applications. Under the architecture, the portions of the application that interact with the user are typically separated from the portions of the application that fulfill client processing requests. Typically, the portions of an application that interact with the user are called a client applications or client software, whereas the portions of the application that service requests made by the client applications are called a server applications or server software. In a network environment, the client applications and server applications are generally executed on different computers.
Historically, digital networks in the form of local area networks, a physical collection of personal computers interconnected with network cabling and network interface cards, consisted of a single network server and multiple network clients. To manage which network clients could access the network server, as well as what files, printers, printer queues, and server applications were available to the network clients, the network server maintained information on each of the resources that were attached to the server and the identities of the network clients and users who could use the services of the network server and the scope and nature of the services available to the network clients and users.
As local area networks became more popular, networks grew in size requiring several servers to service the needs of users. With increased size and complexity of networks, came the need for easier management of network servers. Users required access to an increasing number of services that were located on an increasing number of network servers. Several vendors began offering networking servers. Each vendor implemented a different scheme of providing networking services information. In addition, each network server, because of the way the server maintained information about only its networking services still required management of its resources independently of other network servers.
This insular method of maintaining information of networking services fueled research and development of distributed networking directories, databases that spanned networking servers. Thus far, research has resulted in several potential solutions. Three technologies currently hold greater promise for replacing the large number of insular, idiosyncratic directories that now litter many an enterprise's numerous local-area networks and electronic-mail systems. One approach exploits the X.500 distributed network information directory services protocol developed as published by the CCIT and Open Systems Interconnect consortium.
However, while the X.500 protocol appears to hold the greatest promise to provide a robust, distributed directory, the X.500 protocol has been slow to gain acceptance. The X.500 protocol has been plagued from the start with management, interoperability and security problems. The X.500 protocol specification describes a technical framework, interoperability requirements and compliance criteria but does not describe specific implementations. Thus many of the details of implementation have been left up to systems providers.
The X.500 protocol specification describes a distributed directory. The directory provides information services to network clients. The information in the directory can be read as well as modified by users who have applicable access rights.
The information stored in the directory is a collection of objects with associated attributes or properties. FIG. 1 shows an object called "Computer" with some associated properties, such as owner, operator, status, etc. The values of the properties are not shown in the figure but an example of a value for "Owner" might be "Fred." Objects in the directory and their names correspond to things that humans relate to when dealing with computers, namely, users, printers, print queues, networks and information. Objects such as countries, organizations, networks, people and computers are objects you might find in the directory as well.
The directory provides information to users by giving users a hierarchical view of all of the information contained in the directory. The hierarchical view is generally in the form of a tree. FIG. 2 shows a directory. Each of the branches and terminating points or leaves represent objects in the directory. Generally, implementations of the directory organize objects in subtrees, partitions or domains. FIG. 2 also shows the directory organized into partitions or domains. Multiple copies of each partition may be stored in the directory. Software schemas define and determine the number and types of replicas of each partition.
Multiple replicas of a partition are needed to reduce network storage and traffic requirements and speed up directory searches. Replicas are stored in name servers. A name server is a computer in the network, usually a network server. More than one partition can be stored in a name server. Partitions stored in a name server need not be contiguous.
The directory tree provides a logical means of searching for information. The tree is generally patterned after logical groupings such as organizations, organizational units, computers and users. These logical groupings, while extremely useful in helping users find relevant information also creates significant problems in managing the directory.
Each partition forms a major subtree of the directory. Taken together, the partitions form a hierarchical tree of partitions that leads back to a root partition containing the root directory. Where boundaries of two partitions meet, the partition closer to the root is considered superior, and the partition farther from the root is considered subordinate. Thus, FIG. 2, partitions E and C are subordinate to the other partitions.
The present invention solves one of those problems. As objects of the directory change, the directory must be changed as well. Organizations, organizational units, computers and uses all move. Today, the legal department may be reporting through the finance department. Tomorrow, one of the employees of the finance department might be moved to marketing. Prior to the invention, systems administrators responsible for maintaining a directory had to move each object in the directory in response to a real change in the status of the object. Unfortunately, no facilities existed for moving course grain objects such as an entire department. If the legal department was to be moved to report to the finance department, each object in the legal subtree had to be moved separately.
Summary of the Invention
With the present invention any portion of a directory tree provided that it is at the end of the tree may be moved either within a directory or to another directory. With the invention, ease of providing administration of distributed network directories increases. Accordingly, use of distributed network directories will also increase, making pervasive network computing possible.
Brief Description of the Drawings
The present invention may be more fully understood by reference to the following Detailed Description in conjunction with the Drawings, in which:
FIG. 1 shows a typical directory object, a computer, with some of its associated attributes;
FIG. 2 shows a typical directory tree;
FIG. 3 shows a hypothetical scheme of replicas
FIG. 4 shows how external references use back links.
FIG. 5 shows how the NDS protocol fits within the family of NetWare communications protocols.
FIG. 6 shows the packet structure of NDS packets.
FIG. 7 shows the sequences employed by the invention to move a leaf object.
FIG. 8 shows the movement of a subtree.
Detailed Description of the Invention
The present embodiment of the invention, Novell's NetWare Directory Service or NDS supports moving a terminal or leaf object or partition to allow reorganizing a subtree. Any portion of a directory tree provided that it is at the end of the tree may be moved either within a directory or to another directory.
NDS is based on the X.500 standard and implemented within Novell's NetWare network operating system. Novell implementation of NDS is based on the X.500 standard specification. The X.500 specification does not provide all of the details necessary to implement a distributed network directory.
NDS is implemented within the NetWare network operating system in terms of Novell's native NetWare protocols and a new native protocol called the NDS Protocol. The other components of the native protocols implemented in the NetWare network operating system are illustrated in FIG. 5. IPX is NetWare's native network layer protocol. It provides end-to-end, best-effort datagram delivery. It conveys traffic across a local area network (LAN), a wide area network (WAN), or any internetwork of connected WAN and LAN data-links of like or unlike kinds. An IPX internetwork address has three parts, a Network Number (four bytes) which identifies a network segment, a Node Number (six bytes) which identifies a computer on that segment, and a Socket Number (two bytes) which identifies a software entity within the computer. As an alternative to IPX, NetWare protocols can operate over IP.
The RIP (Routing Information Protocol) supports forwarding of IPX packets over an internetwork. Routers are the devices that connect network segments together. Traditionally, IPX routers exchange connectivity information with each other using RIP to determine proper paths for data packets to take. RIP uses a periodic broadcast mechanism.
The SAP (Service Advertising Protocol) is similar in concept to RIP, but the information exchanged describes services and their addresses (rather than network connectivity). Routers disseminate SAP information as they do with RIP. Through SAP, clients can rendezvous with servers of many kinds. SAP is used to bootstrap a client into the NDS world: a client uses SAP to locate its first NDS server when initializing.
NLSP (NetWare Link Services Protocol) is a newer routing protocol designed to scale to larger environments than RIP and SAP. It plays the role of RIP and SAP for large internetworks. It conveys the same information as RIP and SAP, but instead of using periodic broadcast it sends updates when changes occur.
NCP (NetWare Core Protocol) implements NetWare services such as remote file access. As shown in FIG. 5, NDS packets ride inside NCP packets. While most of this description deals with NDS messages, implementing NDS also involves a few new NCP verbs.
FIG. 6 shows the structure of NDS packets with NCP packets. The data-link header and trailer are media-specific, and are documented in the standard for each LAN or WAN technology. The message formats in this description define an offset for each field. The offset is from the start of the NCP portion of the packet. The first byte of the NCP header is at offset zero. The request/reply fragmentation header is the means of conveying an NDS request and response in a series of NCP request/response exchanges. The arrangement of fields in the NDS data portion of the packet varies, depending on the verb field of the NDS header. Later sections of this document specify the packet formats in detail.
A Completion Code field exists for NCP replies and for NDS replies that ride (fragmented) within NCP. In both cases, Completion Code=0 means "success."Nonzero values report various error conditions.
Moving a Leaf Object
From a network client vantage point, moving a subtree looks the same as moving a single leaf object. From a server standpoint moving a subtree and moving a single leaf object are quite different. The details of moving a leaf object are provided below. The details of moving an entire subtree, and the move from the vantage point of a server, are considered below. See "Moving a Subtree" below.
When a NDS network client moves an NDS directory entry from one container object to another in the tree, it is possible that the source and the destination container objects are in different partitions. A container object is any object that can hold another object, such as a subtree. When a network client moves the NDS entry, there might not be any servers holding a writable replica of both partitions. Each partition can have replicas on several servers. Suppose there are three NDS servers for a name tree: servers S, T, and U. One possible deployment of replicas for FIG. 2 among the servers is illustrated in FIG. 3. There are no restrictions on the placement of replicas together on servers; for example, the replicas stored together on a server need not represent contiguous partitions.
Sometimes, a name server has to keep information about objects that can be outside the replicas it holds. The best examples are the objects superior to its replicas in the tree. Consider FIG. 4. The objects A and Root are not in a replica of a partition on server T. But to know the name of objects in Q, T needs the name of A. This information is kept in a data structure called an external reference. Since an external reference is not in a replica, it is not synchronized with other servers. However, it does have an entry identification valid on the server where it resides (T in this case). If A's name changes, the external reference has to be updated. To allow this, the object A has a back link attribute value pointing to each external reference to A. This is the dotted line in FIG. 4.
Because there might not be any servers holding a writable replica of both partitions when a client attempts to move an NDS entry, moving an object involves two operations: "Begin Move Entry" and "Finish Move Entry."
The detailed steps of moving a leaf object are:
1. The client identifies two servers. The source server is the one holding the master replica of the object being moved. The destination server is the one holding the master replica of the container into which the object is being moved. Sometimes, the servers are the same, but the same procedure applies regardless.
2. The client sends a "Begin Move Entry" request to the destination server. At this point, the destination server enforces access control to permit or deny the operation. The "Begin Move Entry" NDS Protocol verb has the structure identified in Table 1.
3. If no anomalies are detected, the destination server replies with Success. At the same time it records the details of the move operation and starts a ten-minute timer. If the timer expires before the operation completes, it purges its record of the move and step 5 will not complete successfully.
4. The client makes a Finish Move Entry request to the source server. The Finish
Move Entry NDS Protocol verb has the structure identified in Table 2.
5. The source server makes a Restore Entry request to the destination server to transfer the complete object information. This can take several iterations. If there is a temporary anomaly, this step is retried several times before completing or being abandoned. The structure of the Restore Entry NDS Protocol verb is provided in Table 3.
6. If step 5 was successful, the source server removes the entry from its active database. It creates a moved obituary for the entry, identifying the destination location. The obituary propagates to replicas of the source partition through the synchronization channel.
7. The source server sends a Finish Move Entry reply to the client. FIG. 7 illustrates the three-party exchange. The additional steps that follow show the interaction of a wider group of network servers.
8. If another server has an external reference to the old copy of the moved object, the source server holds a Back Link attribute for the object identifying the other server. Using information in the Back Link, it notifies the other server to update the external reference.
9. This uses the Synch External Reference operation. The source uses a "Back Link . . . Moved" obituary for each other server to keep track of which ones have been notified. If new back links appear while this operation progresses, corresponding "Back Link . . . Moved" obituaries are created. The structure of the Synch External Reference NDS Protocol verb is provided in Table 4.
10. Meanwhile, starting at step 3, the destination object has an Inhibit Move obituary attribute attached, indicating that a move is under way. As long as this attribute exists, the object cannot be moved again or deleted. This prevents race conditions as things settle down. Replica synchronization propagates the new object (with its Inhibit Move obituary) throughout replicas of the destination partition.
11. When (a) the deletion of the source object has been propagated throughout the source partition, and (b) the notifications of step 6 have been completed, the object is about to be purged. The source server notifies the destination server using the Release Moved Entry operation. At this point, the destination server removes the Inhibit Move obituary attribute from the new object. Through replica synchronization, the attribute removal propagates to other replicas of the destination partition. When this has occurred, and the destination server purges the obituary, the moved object becomes eligible to be moved again.
Moving a Subtree
As indicated above, from the client's viewpoint, moving a subtree looks the same as moving a single entry. The same Begin Move Entry and Finish Move Entry operations apply, as illustrated in FIG. 3. The exchange among servers is quite different, however.
FIG. 8 shows the move of a subtree. In the example, partition C is being moved under object G. (As customary, the partition is named by its root-most object.) G is in partition F. Partition A is the parent of partition C.
Three partitions participate in the operation. Each has a master replica. In the following detailed discussion of the operation to move a subtree the following terminology is used. The discussions assumes that some of the three partitions can be the same, that some of the servers can be the same, and that:
S is the server holding the master replica of partition A.
T is the server holding the master replica of partition C.
U is the server holding the master replica of partition F.
V is the server holding the master replica of server U's object.
1. The client sends a Begin Move Entry request to U. U enforces its access control to permit or deny the move operation.
2. If all is well, U replies with Success. At the same time, it records the details of the operation and starts a ten-minute timer. If the timer expires before T responds, U purges its record of the move details and the operation will not complete.
3. The client sends a Finish Move Entry request to T. T enforces its access control and the directory schema rules. Also, T locates the object representing server U in the name space and identifies the server, V, holding the master replica of U's object. It sends V a "Control . . . Get Entry Move State" request to determine if U's object is itself moving. If it is moving, the subtree move operation cannot proceed. If any of these checks reveal a problem, T sends an error reply to the client and the operation terminates. The structure of the Control . . . Get Entry Move State NDS Protocol verb is provided in Table 5.
4. T sends a Start Move Tree request to U. U checks the request against its expected details. It also checks that its software version--and the versions of servers identified in the back links of the destination partition root object (F)--are high enough that they support moving a subtree. If all is well, it sends a Success reply to T. In the reply, the Partition Overlap flag is set if partitions A and F are the same partition. The structure of the Start Move Tree NDS Protocol verb is provided in Table 6.
5. U sets F's partition operation to Move Subtree Destination. It sets the partition state and the replica states to Move State O, and the Partition Control Distinguished Name to identify C. If the leaf name of the object is being changed in the course of the move, it also adds a Tree 01d RDN obituary recording the prior name. (With this information, a server can do efficient lookups even if packets arrive from not-yet synchronized servers using an unexpected name.) It starts propagating these changes to the replicas of partition F.
6. T sets C's partition operation to Move Subtree Source. It sets the replica states to Move State O. It also creates three partition control attributes. Each of the three has State=Moved State O and Operation=Move Subtree Source. The Distinguished Name depends on the Type, as follows:
It starts propagating these changes to the replicas of partition C.
7. If the leaf name (relative distinguished name) of the object is being changed in the course of the move, it also adds a Tree New RDN obituary recording the new name. (With this information, a server can do efficient lookups even if packets arrive from not-yet synchronized servers using an unexpected name.) T makes a list of servers to be notified of the operation. The following servers are included in the list (duplicates are suppressed):
Servers holding replicas of partition C.
Servers holding replicas of partition F.
Servers holding external references to objects in C (as identified by back links on the objects).
This is the "Notification List." It is recorded as a Move Subtree obituary for each server on the list. T starts propagating all these changes to the replicas of partition C.
8. If the Partition Overlap flag was not set in step 4, T sends a Control request to S, indicating Lock Partition for partition A (C's parent). This prevents other partition operations on partition A while the move is in progress. For moves within the same partition, it is unnecessary to lock the parent.
9. T sends a Finish Move Entry reply to the client. The client is out of the picture from this point onward.
10. T drives completion of the operation. It sends a Move Tree request to every server in the Notification List, driven by the secondary obituaries. The structure of the Move Tree NDS Protocol request is provided in Table 7.
It persists with periodic retries until all have been contacted. As each server is contacted successfully, T sets the corresponding obituary's Notified flag. The request conveys:
The Tuned Name of C (the source)
The Tuned Name of 6 (the destination)
The Replica Pointer for T (the master server of partition C)
11. T adds a moved obituary to its entry for C, so that any requests about an object in partition C using its old name can be treated correctly while the operation is in progress. When a server, W, on the Notification List receives the request, its action depends on what replicas it holds of partitions C and F. In the following table:
In Cases 1, 2, 3 and 5: W locally switches its record of C's parent from B to G.
In Cases 4 and 7: W locally creates a subordinate reference for partition C. Its reply has the Created Subordinate Reference flag set, informing T to add the subordinate reference to C's replica list.
In Case 6: W locally creates an external reference for G. Its reply has the Created External Reference flag set, informing T to create a back link to W.
In Case 8 and 9: These do not occur. Such servers would not be on the Notification List.
12. Once the servers on the Notification List have been successfully contacted, T sends another request to the same Notification List: End Move Subtree. This causes the Moved obituaries to be purged. Every server has seen the new name, so it is no longer necessary to deal with requests that use objects' old names. As each request completes successfully, the corresponding obituary has the Purgeable flag set.
13. Once all the obituaries are marked Purgeable, T sends a Control request to U and (in the non-Partition Overlap case) to S, indicating Unlock Partition for A and F (respectively). A server receiving this request sets the partition state and the replicas' states to On, and propagates the change to the partition's replicas through the synchronization channel. Finally, T performs the Unlock Partition operation itself for C.
With the present invention any portion of a directory tree provided that it is at the end of the tree may be moved either within a directory or to another directory. With the invention, ease of providing administration of distributed network directories increases. Accordingly, use of distributed network directories will also increase, making pervasive network computing possible.
Although one embodiment of the invention has been illustrated and described, various modifications and changes may be made by those skilled in the art without departing from the spirit and scope of the invention.
Claims
We claim:
1. A method of moving a partition in a distributed directory operating over a plurality of servers, said directory having a plurality of partitions with one or more objects in each of said partitions, said one or more objects including a root object, at least one of said plurality of servers having a replica of one or more of said plurality of partitions and having a hierarchy of superior and subordinate objects in which at least of one of the root objects is subordinate to a superior object, said method comprising the steps of: (a) identifying a target partition having a root object that is subordinate to a source object; (b) identifying a destination object within a destination partition; (c) requesting a move of the target partition from the source object to the destination object; (d) identifying one or more relevant servers that hold one or more of: (i) a replica of the target partition; (ii) a replica of the destination partition; or (iii) a reference to an object in the target partition; and (e) moving the target partition using at least one of the steps of: (i) changing in each relevant server the subordination of the root object in the target partition from the source object to the destination object if such relevant server has at least one of the following: (A) a replica of the target partition and a replica of the destination partition; (B) a replica of the target partition and a reference to the root object of the destination partition; (C) a replica of the target partition and neither a replica of the destination partition nor a reference to the root object of the destination partition; or (D) a reference to the root object of the target partition and a reference to the root object of the destination partition; (ii) creating in each relevant server a reference to the root object of the target partition if such relevant server has at least one of the following: (A) a reference to the root object of the target partition and a replica of the destination partition; or (B) neither a replica of the target partition nor a reference to the root object of the target partition and a replica of the destination partition; or (iii) creating in each relevant server a reference to the destination object if such relevant server has a reference to the root object of the target partition and neither a replica of the destination partition nor a reference to the root object of the destination partition.
2. A method as recited in claim 1 wherein the step of identifying a target partition includes identifying the root object of said target partition which is subordinate to a source object in another partition.
3. A method as recited in claim 1 further including the step of checking that the requested move is permitted in the distributed directory.
4. A method as recited in claim 3 wherein the step of moving the target partition is terminated if it is not completed within a predetermined time period.
5. A method as recited in claim 4 further including the step of starting a timer prior to move is permitted, said predetermined time period being measured by said timer.
6. A method as recited in claim 3 wherein the checking step includes checking that the request to move does not conflict with an access control within the distributed directory, and terminating the move if a conflict exists.
7. A method as recited in claim 3 wherein the checking step includes checking that the requested move does not conflict with directory schema rules within the distributed directory, and terminating the move if such a conflict exists.
8. A method as recited in claim 3 wherein the checking step includes checking that the requested move does not conflict with another operation involving the target partition, and terminating the move if such conflict exists.
9. A method as recited in claim 3 wherein the checking step includes checking that the relevant servers can support the requested move, and terminating the move if the relevant servers cannot support the requested change.
10. A method as recited in claim 3 further comprising the step of locking the replicas of the target partition and the destination partition until the move is complete.
11. A method as recited in claim 1 further comprising the step of locking replicas of the partition holding the source object if said source object partition overlaps the target partition until the move is complete.
12. A method as recited in claim 1 further providing to one or more of said relevant servers (i) the identity of the root directory; and (ii) the identity of the destination object.
13. A method of changing the subordination of a partition in a distributed directory operating over a plurality of servers, said directory having a plurality of partitions with one or more objects in each of said partitions, said one or more objects including a root object, at least one of said plurality of servers having a replica of one or more of said plurality of partitions and having a hierarchy of superior and subordinate objects in which at least of one of the root objects is subordinate to a superior object, said method comprising the steps of: (a) identifying a target partition having a root object that is subordinate to a source object; (b) identifying a destination object within a destination partition; (c) requesting a change in the subordination of the root object of the target partition from the source object to the destination object; (d) identifying one or more relevant servers that hold one or more of: (i) a replica of the target partition; (ii) a replica of the destination partition; or (iii) a reference to an object in the target partition; (e) changing the subordination of the target partition by changing in each relevant server the subordination of the root object in the target partition from the source object to the destination object if such relevant server has at least one of the following: (i) a replica of the target partition and a replica of the destination partition; (ii) a replica of the target partition and a reference to the root object of the destination partition; (iii) a replica of the target partition and neither a replica of the destination partition nor a reference to the root object of the destination partition; or (iv) a reference to the root object of the target partition and a reference to the root object of the destination partition.
14. A method as recited in claim 13 further including the step of providing to one or more of said relevant servers (i) the identity of the root directory and (ii) the identity of the destination object.
15. A method as recited in claim 13 further including the step of checking that the requested move is permitted in the distributed directory.
16. A method of changing the subordination of a partition in a distributed directory operating over a plurality of servers, said directory having a plurality of partitions with one or more objects in each of said partitions, said one or more objects including a root object, at least one of said plurality of servers having a replica of one or more of said plurality of partitions and having a hierarchy of superior and subordinate objects in which at least of one of the root objects is subordinate to a superior object, said method comprising the steps of: (a) identifying a target partition having a root object that is subordinate to a source object; (b) identifying a destination object within a destination partition; (c) requesting a change in the subordination of the root object of the target partition from the source object to the destination object; (d) identifying one or more relevant servers that hold one or more of: (i) a replica of the target partition; (ii) a replica of the destination partition; or (iii) a reference to an object in the target partition; (e) changing the subordination of the target partition by creating in each relevant server a reference to the root object of the target partition if such relevant server has at least one of the following: (i) a reference to the root object of the target partition and a replica of the destination partition; or (ii) neither a replica of the target partition nor a reference to the root object of the target partition and a replica of the destination partition.
17. A method as recited in claim 16 further including the step of providing to one or more of said relevant servers (i) the identity of the root directory and (ii) the identity of the destination object.
18. A method as recited in claim 16 further including the step of checking that the requested move is permitted in the distributed directory.
19. A method of changing the subordination of a partition in a distributed directory operating over a plurality of servers, said directory having a plurality of partitions with one or more objects in each of said partitions, said one or more objects including a root object, at least one of said plurality of servers having a replica of one or more of said plurality of partitions and having a hierarchy of superior and subordinate objects in which at least of one of the root objects is subordinate to a superior object, said method comprising the steps of: (a) identifying a target partition having a root object that is subordinate to a source object; (b) identifying a destination object within a destination partition; (c) requesting a change in the subordination of the root object of the target partition from the source object to the destination object; (d) identifying one or more relevant servers that hold one or more of: (i) a replica of the target partition; (ii) a replica of the destination partition; or (iii) a reference to an object in the target partition; (e) changing the subordination of the target partition by creating in each relevant server a reference to the destination object if such relevant server has a reference to the root object of the target partition and neither a replica of the destination partition nor a reference to the root object of the destination partition.
20. A method as recited in claim 19 further including the step of providing to one or more of said relevant servers (i) the identity of the root directory and (ii) the identity of the destination object.
21. A method as recited in claim 19 further including the step of checking that the requested move is permitted in the distributed directory.
Patent Citations (6)
Non-Patent Literature (8)
- Accessing Files in an Internet: The Jade File System, Herman C. Rao and Larry L. Peterson. 8198 IEEE Transaction on Software Engineering 19(1993) Jun., No. 6, New York, US.
- Exploiting Read-Mostly Workoads in The FileNet File System, David A. Edwards and Martin S. McKendry, 8283 Operating Systems Review (SIGOPS) 23 (1989) No. 5 New York, US.
- Elias Drakopoulos and Matt Merges, "Performance study of client-server storage system.", IEEE, 1991, pp. 67-72.
- Bhide et al. "A comparison of two Approaches to build reliable distributed file servers.", IEEE, 1991, pp. 616-623.
- Mutka et al. "Managing personal files across independent file management units", IEEE, 1992, pp.254-261.
- R. T. Hurley et al. "Limited effects of finite storage on a beneficial file migration policy."; IEEE, 1994, pp. 432-439.
- Elias Drakopoulos and Matt Merges, "Performance study of client-server storage system.", IEEE, 1992, vol. 41; No. 11; pp. 1442-1452.
- Robert Cowart, "Mastering Windows 3.1"; 1992, SYBEX Inc.; pp. 132-141.