Efficient distributed cache consistency

• US 9,262,324 B2
• Filed: 06/12/2014
• Issued: 02/16/2016
• Est. Priority Date: 12/12/2013
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Fried, Eric P., Russell, Lance W.
• Primary Examiner(s): YU, JAE UN
• Application Number:
  US14/302,981
• Publication Number:
  US 20150169497A1
• Time in Patent Office: 614 Days
• Field of Search: 711/141
• US Class Current: 1/1
• CPC Class Codes: G06F 12/0822 Copy directories local copy tags for implementing a bus snooping protocol G06F12/0831
• CPC Class Codes: G06F 12/0831 using a bus scheme, e.g. with bus monitoring or watching means
• CPC Class Codes: G06F 12/084 with a shared cache
• CPC Class Codes: G06F 15/17331 Distributed shared memory [DSM], e.g. remote direct memory access [RDMA]
• CPC Class Codes: G06F 2212/154 Networked environment
• CPC Class Codes: G06F 2212/60 Details of cache memory

Abstract

Mechanisms are provided for performing a write operation on a shared resource in a cluster of data processing systems. The mechanisms determine whether a locally cached copy of the shared resource is present in a local cache memory. The mechanisms, in response to a determination that a locally cached copy is present, determine whether the locally cached copy is a latest version of the shared resource. The mechanisms, in response to determining that that locally cached copy is a latest version: perform the write operation on the locally cached copy to generate an updated locally cached copy, and transmit a cluster event notification to other data processing systems of the cluster indicating the shared resource was written to and which data processing system performed the write operation. The cluster event notification is logged in at least one change log data structure associated with the other data processing systems.

Claims

1. A method, in a data processing system comprising a processor and a memory, for performing a write operation on a shared resource in a cluster of data processing systems, the method comprising:

determining, by the data processing system, whether a locally cached copy of the shared resource is present in a local cache memory of the data processing system;

in response to a determination that a locally cached copy of the shared resource is present in the local cache memory, determining whether the locally cached copy of the shared resource is a latest version of the shared resource; and

in response to determining that that locally cached copy of the shared resource is a latest version of the shared resource;

performing the write operation on the locally cached copy of the shared resource to generate an updated locally cached copy of the shared resource; and

transmitting a cluster event notification to other data processing systems of the cluster indicating the shared resource was written to and identifying which data processing system in the cluster performed the write operation, wherein the cluster event notification is logged in at least one change log data structure associated with the other data processing systems.

2. The method of claim 1, wherein determining whether a locally cached copy of the shared resource is a latest version of the shared resource comprises:

searching a change log data structure associated with the data processing system for an entry corresponding to the shared resource, wherein the change log data structure stores entries corresponding to cluster event notifications received from other data processing systems in the cluster in response to the other data processing systems accessing shared resources of the cluster; and

determining, based on results of the searching of the change log data structure, whether entries in the change log data structure indicate that the locally cached copy of the shared resource is a latest version of the shared resource.

3. The method of claim 2, wherein determining whether entries in the change log data structure indicate that the locally cached copy of the shared resource is a latest version of the shared resource comprises:

identifying a latest entry in the change log data structure corresponding to the shared resource based on a staleness indicator of the entries in the change log data structure; and

determining if a source identifier in the latest entry matches an identifier of the data processing system, wherein the source identifier identifies a data processing system that performed a write operation to the shared resource.

4. The method of claim 2, wherein determining whether entries in the change log data structure indicate that the locally cached copy of the shared resource is a latest version of the shared resource comprises:

determining whether there are no entries corresponding to the shared resource in the change log data structure; and

in response to determining that there are no entries corresponding to the shared resource in the change log data structure, determining that the locally cached copy of the shared resource is not the latest version of the shared resource.

5. The method of claim 2, further comprising:

in response to determining, based on the entries in the change log data structure, that the locally cached copy of the shared resource is not the latest version of the shared resource, identifying a storage location of the latest version of the shared resource based on the entries in the change log data structure;

reading a the latest version of the shared resource from the storage location of the latest version of the shared resource; and

storing a new locally cached copy of the shared resource in the local cache memory of the data processing system.

6. The method of claim 1, wherein the cluster event notification comprises at least one of a version number of the updated locally cached copy of the shared resource or a timestamp corresponding to a time when the write operation was performed on the locally cached copy of the shared resource, a unique identifier of the data processing system, and a unique identifier of the shared resource.

7. The method of claim 1, further comprising performing a flush of the local cache memory of the data processing system in response to an event, wherein performing the flush of the local cache memory comprises:

determining if a locally cached version of the shared resource is the latest version of the shared resource; and

based on results of determining if the locally cached version of the shared resource is the latest version of the shared resource, either writing the locally cached version back to a source storage location of the shared resource or deleting the locally cached copy of the shared resource from the local cache memory.

8. The method of claim 7, wherein if the results of determining if the locally cached version of the shared resource is the latest version of the shared resource indicate that the locally cached version of the shared resource is the latest version of the shared resource, the method comprises writing the locally cached version back to the source storage location of the shared resource and sending an event notification to the other data processing systems of the cluster indicating that the latest version of the shared resource is located at the source storage location.

9. The method of claim 1, wherein the at least one change log data structure comprises a plurality of change log data structures, each change log data structure being resident on a corresponding one of the data processing systems in the cluster.

10. The method of claim 1, wherein the at least one change log data structure comprises a single change log data structure associated with a source storage location of the shared resource.