Multiprocessing computer system employing local and global address spaces and COMA and NUMA access modes

• US 5,887,138 A
• Filed: 07/01/1996
• Issued: 03/23/1999
• 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., Loewenstein, Paul N.
• Primary Examiner(s): Lee, Thomas C.
• Assistant Examiner(s): PATEL, GAUTAM
• Application Number:
  US08/675,635
• Time in Patent Office: 995 Days
• Field of Search: 395/800.28, 395/800.27, 395/200.43, 395/200.45, 395/200.46, 395/200.42, 395/200.44, 395/475, 395/480, 395/468, 395/800.1, 711/148, 711/141
• US Class Current: 709/215
• CPC Class Codes: G06F 12/0284 Multiple user address space allocation, e.g. using different base addresses interprocessor communication G06F15/163
• CPC Class Codes: G06F 12/0813 with a network or matrix configuration
• CPC Class Codes: G06F 12/0817 using directory methods
• CPC Class Codes: G06F 2212/2542 Non-uniform memory access [NUMA] architecture
• CPC Class Codes: G06F 2212/272 Cache only memory architecture [COMA]

Abstract

A multiprocessing computer system employs local and global address spaces and Non- Uniform Memory Architecture (NUMA) and Cache-Only Memory Architecture (COMA) access modes. The multiprocessing computer architecture employs a plurality of processing nodes. When a processing node initiates a memory transaction, the node determines whether the address of the memory transaction is a global address or a local physical address. If the address is a global address, a NUMA coherency request is initiated. Alternatively, if the address is a local physical address, a COMA coherency request is initiated. The nodes additionally include local physical address to global address translation units. The local physical address to global address translation units are configured to translate a local physical address to a corresponding global address prior to initiating a COMA coherency request.

Claims

1. A multiprocessing computer system comprising:

a first processing node including a first processor, a first memory, and a first system interface; and

a second processing node coupled to said first processing node, said second processing node including a second memory, wherein said first memory and said second memory comprise a distributed shared memory system;

wherein said first processor is configured to initiate a first transaction having a first address, wherein a first portion of said first address is indicative of a location of a coherency unit corresponding to said first address within said distributed shared memory system, wherein said coherency unit is a number of contiguous bytes of memory,

wherein a COMA mode is selected for said first transaction if said first portion of said first address indicates that said coherency unit is stored in said first memory, and wherein a NUMA mode is selected for said first transaction if said first portion of said first address indicates that said coherency unit is stored in said second memory, and

wherein said first system interface is configured to initiate a NUMA coherency request responsive to said first transaction and said NUMA mode being selected for said first transaction, and wherein said first system interface is configured to initiate a COMA coherency request responsive to said first transaction, said COMA mode being selected and a corresponding coherency unit stored within said first memory is a copy of a third coherency unit stored within said second memory, and wherein said NUMA coherency request causes said first processor to complete said first transaction upon data stored within said second memory, wherein said COMA coherency request causes said first processor to complete said first transaction upon data stored within said first memory.

2. The multiprocessing computer system as recited in claim 1 wherein said first system interface comprising a translation unit configured to translate said first address from a local physical address specifying said first coherency unit to a corresponding global address specifying said third coherency unit prior to initiating said COMA coherency request.

3. The multiprocessing computer system as recited in claim 2 wherein said first system interface further comprises a storage for storing a plurality of coherency states including a coherency state corresponding to said local physical address.

4. The multiprocessing computer system as recited in claim 3 wherein said first system interface is configured to determine if said COMA coherency request is to be generated by examining said coherency state.

5. The multiprocessing computer system as recited in claim 3 wherein said plurality of coherency states includes a coherency state for each coherency unit stored in said first memory.

6. The multiprocessing computer system as recited in claim 1 wherein said first processing node comprises a first plurality of processors including said first processor.

7. The multiprocessing computer system as recited in claim 6 wherein said first processing node comprises a symmetric multiprocessing node.

8. The multiprocessing computer system as recited in claim 7 wherein said first plurality of processors are coupled to provide transactions upon a shared bus within said first processing node, said first system interface also coupled to said shared bus.

9. The multiprocessing computer system as recited in claim 8 wherein each of said first plurality of processors is coupled to said shared bus via a respective one of a first plurality of external caches.

10. The multiprocessing computer system as recited in claim 1 further comprising a third processing node coupled to said first processing node and said second processing node, wherein said second processing node is configured to generate a coherency demand for said third processing node in response to a coherency request from said first processing node if said third processing node is storing a second copy of said third coherency unit, and wherein said third processing node is configured to transmit a coherency reply to said first processing node in response to said coherency demand.

11. The multiprocessing computer system as recited in claim 10 wherein said coherency reply includes said third coherency unit, and wherein said first system interface is configured to provide said third coherency unit to said first processor, and wherein said first system interface is further configured to store said third coherency unit as said first coherency unit in said first memory if said coherency request is said COMA coherency request.

12. A system interface for a processing node in a multiprocessing system comprising

a system interface logic unit coupled to receive a transaction initiated by a processor within said processing node, said system interface logic unit is configured to generate a COMA coherency request in response to said transaction if a portion of an address corresponding to said transaction indicates a COMA mode for said transaction and a corresponding coherency unit stored within said first memory is a copy of a third coherency unit stored within said second memory, wherein said COMA coherency request causes said processor to complete said transaction upon data stored within a memory within said processing node, and wherein said system interface logic unit is configured to generate a NUMA coherency request in response to said transaction if said portion of said address indicates a NUMA mode for said transaction, wherein said NUMA coherency request causes said processor to complete said transaction upon data stored within a memory in a second processing node;

and

a translation unit coupled to said system interface logic unit, wherein said translation unit is configured to translate said address to a corresponding global address, said system interface logic unit is configured to receive said corresponding global address from said translation unit prior to initiating said COMA coherency request.

13. The system interface as recited in claim 12 wherein said system interface logic unit is configured to use said corresponding global address as an address for said COMA coherency request.

14. The system interface as recited in claim 12 farther comprising a storage configured to store a coherency state corresponding to a coherency unit, wherein said coherency unit is a number of contiguous bytes of memory, and wherein said coherency unit corresponds to said address of said transaction if said portion of said address indicates said COMA mode.

15. The system interface as recited in claim 14 further comprising a transaction filter coupled to said storage and said system interface logic unit, wherein said transaction filter is configured to determine if an access right represented by said coherency state is sufficient for said transaction to complete within said processing node, and wherein, if said access right is insufficient for said transaction to complete within said processing node, said transaction filter is configured to convey said transaction to said system interface logic unit, and wherein, if said access right is sufficient for said transaction to complete within said processing node, said transaction filter is configured to inhabit conveyance of said transaction to said system interface logic unit.

16. A method for operating a multiprocessing computer system including a first processing node comprising a first processor and a first memory, said multiprocessing computer system further including a second processing node, the method comprising:

initiating a transaction having an address corresponding to a coherency unit, said initiating performed by said first processor, wherein said coherency unit is a number of contiguous bytes of memory;

generating a COMA coherency request if a portion of said address indicates a COMA mode for said transaction and a corresponding coherency unit stored within said first memory is a copy of a different coherency unit in a second memory within said second processing node, wherein said COMA coherency request causes said first processor to complete said transaction upon data stored within said first memory; and

generating a NUMA coherency request if said portion of said address indicates a NUMA mode for said transaction, wherein said NUMA coherency request causes said first processor to complete said transaction upon data stored within said second memory.

17. The method as recited in claim 16 further comprising receiving a coherency reply in said first processing node, said coherency reply including said coherency unit, and providing said coherency unit to said first processor.

18. The method as recited in claim 17 further comprising storing said coherency unit in said first memory, if said coherency reply is responsive to said COMA coherency request.

19. The method as recited in claim 18 further comprising inhibiting storage of said coherency unit in said first memory, if said coherency reply is responsive to said NUMA request.

20. The method as recited in claim 16 wherein said generating a COMA coherency request is performed if an access right stored in said first processing node is insufficient for said transaction to complete within said first processing node.