Multiprocessing system configured to perform efficient write operations

• US 5,749,095 A
• Filed: 07/01/1996
• Issued: 05/05/1998
• Est. Priority Date: 07/01/1996
• Status: Expired due to Term

• Current Assignee: Oracle America, Inc. (Oracle Corporation)
• Original Assignee: Sun Microsystems Incorporated (Oracle Corporation)
• Inventors: Hagersten, Erik E.
• Primary Examiner(s): Kim, Matthew M.
• Application Number:
  US08/675,634
• Time in Patent Office: 673 Days
• Field of Search: 711/141-146, 711/121, 711/130, 711/3, 711/201, 711/203, 711/117, 711/206-210, 711/140, 395/200.01-200.09, 395/200.14
• US Class Current: 711/141
• CPC Class Codes: G06F 12/0817 using directory methods

Abstract

A computer system defines a "fast write" protocol for performing certain write operations. Write operations include a particular encoding if they are to be performed using the fast write protocol. When the system interface within a node detects the particular encoding, the write operation is captured by the system interface. In addition, the data is transferred to the system interface from the processor performing the write operation. The data transfer is performed even if the node is not maintaining a coherency state for the affected coherency unit which is consistent with performing the write operation. Instead, the coherency activity employed to acquire the proper coherency state is initiated subsequent to or in parallel with the receipt of data from the processor. Because fast write operations are performed prior to acquiring write permission to the coherency unit, ordering with respect to other operations is not maintained. Therefore, the fast write protocol is not suitable for all write operations within the computer system. However, the protocol may be used to increase performance. For example, a group of writes enveloped by software synchronization operations appear to be ordered as a group with respect to operations outside of the synchronization. The performance gained by executing the group of writes using the fast write protocol may outweigh the system bandwidth and extra latency used to perform synchronization.

Claims

1. A method for performing write operations in a multiprocessing computer system, comprising:

initiating a write operation by a processor within a local processing node of said multiprocessing computer system;

performing a coherency operation to at least one remote processing node in response to said write operation;

completing said write operation within said local processing node prior to completion of said coherency operation if said write operation includes a specific predefined encoding; and

completing said write operation within said local processing node subsequent to completion of said coherency operation if said write operation includes an encoding different than said specific predefined encoding.

2. The method as recited in claim 1 wherein said specific predefined encoding is provided via an address included with said write operation.

3. The method as recited in claim 2 wherein said address lies within a first address region which is within an address space of said local processing node.

4. The method as recited in claim 3 wherein said first address region is identified by a particular value within a plurality of most significant bits of said address.

5. The method as recited in claim 3 wherein said first address region is an alias for a second address region within said address space.

6. The method as recited in claim 4 wherein said encoding different than said specific predefined encoding comprises a second address within a second address region.

7. The method as recited in claim 3 wherein said write operation is a write stream operation.

8. The method as recited in claim 3 further comprising translating said address to a global address prior to said performing said coherency operation.

9. The method as recited in claim 1 wherein said completing comprises transferring data from said processor.

10. The method as recited in claim 9 further comprising transferring said data to a home node of said address upon completion of said coherency operations.

11. An apparatus for performing write operations in a multiprocessing computer system, comprising:

a processor configured to perform a write operation; and

a system interface coupled to receive said write operation and to perform a coherency operation in response to said write operation, wherein said system interface is configured to complete said write operation with respect to said processor prior to completing said coherency operation if said write operation includes a specific predefined encoding, and wherein said system interface is further configured to inhibit completion of said write operation with respect to said processor until completion of said coherency operation if said write operation includes a different encoding than said specific predefined encoding.

12. The apparatus as recited in claim 11 wherein said coherency operation is performed in order to acquire a coherency state which allows said write operation to occur to a coherency unit identified by said write operation.

13. The apparatus as recited in claim 11 wherein said specific predefined encoding is provided via an address included with said write operation.

14. The apparatus as recited in claim 13 wherein said address lies within a first address region within an address space of accessible to said processor.

15. The apparatus as recited in claim 14 wherein said first address region is an alias to a second address region within said address space, and wherein said different encoding comprises a second address lying within said second address region.

16. The apparatus as recited in claim 11 wherein completing said write operation with respect to said processor comprises transferring data corresponding to said write operation from said processor.

17. A computer system, comprising:

a first processing node including at least one processor, wherein said processor is configured to perform a write operation, and wherein said first processing node is configured to complete said write operation with respect to said processor prior to acquiring a coherency state allowing said write operation if said write operation includes a predefined encoding; and

a second processing node configured as a home node of a coherency unit affected by said write operation, wherein said second processing node is coupled to receive a coherency request from said first processing node, and wherein said first processing node conveys said coherency request in order to acquire said coherency state.

18. The computer system as recited in claim 17 wherein said predefined encoding comprises an address within an address region of an address space corresponding to said first processing node.

19. The computer system as recited in claim 18 wherein said address region is an alias to a second address region within said address space.

20. The computer system as recited in claim 17 wherein said first processing node provides data corresponding to said write operation to said second processing node upon completion of said coherency request.