• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    In a SCI based multi-node system, the write purge command joins the new node that is requesting to write to the memory of the sharing list, while maintaining the connection between the memory and the sharing list. The new node then issues the purging command to each node in the sharing list, while still maintaining the connection of the sharing list to the memory. Next, the new node issues the collapsing command to separate the sharing list from the memory after the purging command has been issued to each node. A send request data packet is used to distribute the write purge command to the memory node.
    公开号:US20010003838A1
    申请号:US08/792,077
    申请日:1997-01-31
    公开日:2001-06-14
    发明作者:Bryan Hornung；Bryan Marietta
    申请人:Hewlett Packard Co；
    IPC主号:G06F12-0813
    专利说明:
    [0001] This invention relates in general to memory accesses in multi-node, multi-processor, cache coherent non-uniform access system and relates in particular to a system and method for performing a write purge operation in such a system. [0001] BACKGROUND OF THE INVENTION
    [0002] The Scalable Coherent Interface (SCI) Direct Memory Access (DMA) write operations in the standard SCI specification completely over-writes the cache lines. If the DMA devices are only updating some of the bytes in the line, the information contained in the memory for the other bytes are lost in the complete overwrite. [0002]
    [0003] The DMA write operation also relies on software to maintain cache coherency. If two devices are writing the same line, the second device can believe that it has finished purging all caches that contain the old data, even though the lines are being purged because of the first device. This second device can then allow other software to read stale data. The use of stale data by the software will cause program errors. [0003]
    [0004] Other prior art methods for writing partial lines rely on reading the line into a local cache before the specified bytes are updated. This results in other desirable data being swapped out of the cache because of conflicts with the stored read data. In the end, this results in poor performance of other processes that are currently running because of the extra memory operations and the latency associated with refetching affected data. [0004]
    [0005] Therefore, there is a need in the art for a method and system that has a write command that does not allow the use of stale data by the software. [0005]
    [0006] In addition, there is a need in the art for a method and system that does not require the reading memory lines into a local cache before the updating memory bytes. [0006] SUMMARY OF THE INVENTION
    [0007] These and other objects and features are achieved in a system in which follows the same general flow as the DMA Write as described in the SCI specification, however, the inventive system and method does not detach the cache sharing list from memory. Instead, it joins the sharing list. This prevents another write purge from another node from believing it has finished its operation while memory lines are still encached. If no sharing list exists, a mask supplied by the command is used to merge the new data into memory. The system and method tracks down stale data in remote caches and merges it into the memory line using a mask instead of discarding it. [0007]
    [0008] One technical advantage of the present invention is to issue a write purge command that joins the new node to the sharing list, while maintaining the connection between the memory and the sharing list. [0008]
    [0009] Another technical advantage of the present invention is to have the new node issue the purging command to each node in the sharing list, while maintaining the connection between the memory and the sharing list. [0009]
    [0010] A further technical advantage of the present invention is to have the new node issue the collapsing command to separate the sharing list from the memory after the purging command has been issued to each node. The collapsing command completes the destruction of the sharing list. [0010]
    [0011] A further technical advantage of the present invention is to use a write mask with the write purge command. [0011]
    [0012] The foregoing has outlined rather broadly the features and technical advantages of the present invention in order that the detailed description of the invention that follows may be better understood. Additional features and advantages of the invention will be described hereinafter which form the subject of the claims of the invention. It should be appreciated by those skilled in the art that the conception and the specific embodiment disclosed may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present invention. It should also be realized by those skilled in the art that such equivalent constructions do not depart from the spirit and scope of the invention as set forth in the appended claims. [0012] BRIEF DESCRIPTION OF THE DRAWINGS
    [0013] For a more complete understanding of the present invention, and the advantages thereof, reference is now made to the following descriptions taken in conjunction with the accompanying drawings, in which: [0013]
    [0014] FIGS. 1A to [0014] 1C show an example of the SCI specification write purge operation;
    [0015] FIGS. 2A to [0015] 2C show an example of the inventive write purge partial operation;
    [0016] FIG. 3 describes the operations performed by an initialization state machine to execute the write purge operation; [0016]
    [0017] FIG. 4 describes the operations performed by a request state machine to execute the write purge operation; [0017]
    [0018] FIGS. 5A to [0018] 5F describe the operations performed by a response state machine to execute the write purge operation;
    [0019] FIGS. 6A to [0019] 6C describe the operations performed by a conflicting state machine to execute the write purge operation;
    [0020] FIG. 7 depicts the send request data packet for carrying the write purge command in the bytemask; and [0020]
    [0021] FIG. 8 shows the write mask. [0021] DESCRIPTION OF THE PREFERRED EMBODIMENTS
    [0022] FIGS. 1A to [0022] 1C show how the prior art SCI specification performs a DMA write to a line that is fresh with a sharing list. The list starts out with memory 101 in the state of fresh with pointer to node 0 102. Node 0 102 is head_fresh and has a forward pointer to node 1 103, which is mid_valid. Node 1 103 has a back pointer back to the head, node 0 102, and a forward pointer to node 2 104. Node 2 104 is tail_valid and has a back pointer to the mid or node 1 103.
    [0023] The first step that is performed by the standard SCI flow is to determine if another node wants to write this memory line, if so, then it will issue an MWRITE_FRESH_TO_HOME. As shown in FIG. 1A, node [0023] 3 105 wants to write to the memory line and issues the MWRITE command to begin the write operation. FIG. 1B shows that the list fresh_to_home is actually detached from the memory 101 so that the sharing list of nodes 0, 1, and 2 102, 103, 104 is no longer connected to the memory 101. So the memory is now marked home and any new accesses can retrieve the memory line directly from memory 101.
    [0024] Now node [0024] 3 105 goes in and cleans up or purges the sharing list. Node 3 105 sends CREAD00_FRESH_TO_INVALID from node 3 to node 0. When node 3 105 receives a response, it uses the response to get the forward pointer. Node 3 105 then uses the forward pointer to send CREAD00_VALID_TO_INVALID from node 3 105 to node 1 103. When node 3 105 receives a response, it then uses the forward pointer from the response to send the CREAD00_VALID_TO_INVALID from node 3 105 to node 2 104. When this operation is complete or done, only memory in the home state contains a copy of the data.
    [0025] The SCI specification relies on software interlocking in such a way that it does not access the fresh memory until the list is totally purged. This can be very difficult if multiple write purges occur to the same memory line almost simultaneously, as the second one will immediately see that the line is marked HOME and allow software to access this line, which could still have stale copies. [0025]
    [0026] FIGS. 2A to [0026] 2C shows that the operation of the invention commences similarly to that shown in FIG. 1A, beginning with the initial state with memory 101 being marked fresh and its pointer to node 0 102. Node 0 102 is marked head_fresh with a forward pointer to node 1 103 which is mid_valid. Mid_valid has a back pointer to the head and a forward pointer to node 2 104, which is tail_valid. Again, node 3 105 wants to do a write_purge or an update_memory with the new data. So node 3 105 sends an MWRITE_purge command, MWRITE_PU_ATTACH_TO_LIST, from node 3 105 to memory 101. As shown in FIG. 2B, memory 101 and node 3 105 uses the mask 500 from the write_purge_partial command to merge in the new data and does not detach the list as in FIG. 1B.
    [0027] So now the list is still connected with the memory [0027] 101 marked fresh. The forward pointer of the memory 101 points to node 3 105. Node 3 105 points to node 0 102. Node 0 102 remains in the head_fresh state since it has not received any communication from node 3 105. Node 0 102 has a forward pointer pointing to node 1 103 mid_valid, and the mid_valid has a forward pointer to tail_valid node 2 104.
    [0028] After node [0028] 3 105 receives its response from the mwrite_purge command, it then does a CPURGE not a CREAD, because in this case the line is FRESH and does not need new data. So after the mwrite_purge response, node 3 105 starts purging the list with a CPURGE_FRESH_TO_INVALID from node 3 105 to node 0 102. Node 3 105 then receives a response from node 0 containing the next forward pointer, and issues a CPURGE_VALID_TO_INVALID from node 3 105 to node 1 103.
    [0029] When node [0029] 3 105 receives a response from node 1 103 with the next forward pointer, it then performs a CPURGE_VALID_TO_INVALID from node 3 105 to node 2 104. When node 3 105 sees that the tail_valid node, here node 2 104, has been reached, it then does an MUPDATE_LIST_TO_HOME from node 3 105 to the memory 101 and the memory is left in the home state as shown in FIG. 2C. Thus, this method achieves the same results as shown in FIG. 1C.
    [0030] FIGS. [0030] 3 to 6 describe the operations handled by the various state machines inside the SCI controller to do the write_purge_partial flow.
    [0031] FIG. 3 describes the operations for the memory access controller Request State Machine Logic. This state machine will take requests from the memory access controller, and given the state passed by the memory access controller and the flow, will decide on the first state that is inside the SCI controller. For example, in the write_purge case, where the memory access controller requests with head_fresh, the SCI controller would initialize this request to a CS_HF_MODS_HD state, or for in a different case, head_fresh_MODS_only_head_dirty state. [0031]
    [0032] FIG. 4 describes how that state is turned into a request on the ring. In the previous case where the state is CS_HF_MODS_HD, this state machine would then make a ring request of mupdate_list_to_gone. [0032]
    [0033] FIGS. [0033] 5A-F describes the operations for the Response State Machine. This state machine describes what is done when a response is received from the remote memory or cache. The machine decides whether to make another request to the ring, make a response to the memory access controller, or in some cases, perform both. Again following the previous case where CS_HF_MODS_HD, if the response is not nullified and is FRESH, in that case, then the machine will transition to the CS_HD_INVAL_OD state and then the FIG. 3B state machine will take this state and generate a new request.
    [0034] The state machine of FIGS. [0034] 6A-C describes what happens if a request for the same line is received from another node while actively in the write purge flow. Again following the previous case where CS_HF_MODS_HD state, the state machine will allow a cupdate_prevmid or cupdate_prevtail to complete. All other requests will be nullified. The prev-TAIL will update the forward pointer and change the state.
    [0035] FIG. 7 depicts the symbols in a send request packet [0035] 400 on the SCI rings. The significant field in this particular packet is the second symbol in which we have a field labeled bytemask 401. The bytemask field 401 is used to carry the byte mask information for the write_purge_partial.
    [0036] FIG. 8 describes the mask [0036] 500 in more detail. The mask consists of a start 501 and an end 502, each being 5 bits. If the start is all 0's and the end is all 1's that means we had started at 0 and ended at 31, meaning we write the whole line per write_purge. Other values may be used, for example, start at 5 and end at 20, however the sub-field must always be of continuous bytes.
    [0037] Although the present invention and its advantages have been described in detail, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims. [0037]
    权利要求:
    Claims (28)
    [1" id="US-20010003838-A1-CLM-00001] 1. A method for performing a write operation on a memory with a sharing list in a multi-node system, the method comprising the steps of:
    issuing a write linking command that attaches a new node to the sharing list; and
    maintaining a connection between the memory and the sharing list while the new node is attaching to the sharing list.
    [2" id="US-20010003838-A1-CLM-00002] 2. The method according to
    claim 1 , wherein:
    the linking command is issued by the new node.
    [3" id="US-20010003838-A1-CLM-00003] 3. The method according to
    claim 1 , wherein:
    the sharing list, the new node, and the memory are connected by a plurality of pointers.
    [4" id="US-20010003838-A1-CLM-00004] 4. The method according to
    claim 1 , wherein:
    the new node is connected between the memory and the shared list.
    [5" id="US-20010003838-A1-CLM-00005] 5. The method according to
    claim 1 , wherein:
    the connection is between the new node and the memory.
    [6" id="US-20010003838-A1-CLM-00006] 6. The method according to
    claim 1 , further comprising the steps of:
    issuing a purging command to each node in the sharing list; and
    issuing a collapsing command to separate the sharing list from the memory after the purging command has been issued to each node.
    [7" id="US-20010003838-A1-CLM-00007] 7. The method according to
    claim 6 , wherein:
    the purge command invalidates the data stored at each node.
    [8" id="US-20010003838-A1-CLM-00008] 8. The method according to
    claim 6 , wherein:
    the purge command is issued by the new node.
    [9" id="US-20010003838-A1-CLM-00009] 9. The method according to
    claim 6 , wherein:
    the step of issuing the collapsing command also separates each node in the sharing list from each other.
    [10" id="US-20010003838-A1-CLM-00010] 10. The method according to
    claim 6 , wherein:
    the step of issuing the collapsing command also separates the new node from the memory.
    [11" id="US-20010003838-A1-CLM-00011] 11. The method according to
    claim 6 , wherein:
    the collapsing command is issued by the new node.
    [12" id="US-20010003838-A1-CLM-00012] 12. The method according to
    claim 1 , further comprising the step of:
    using a data packet to distribute the command for the write operation throughout the system.
    [13" id="US-20010003838-A1-CLM-00013] 13. The method according to
    claim 1 , further comprising the step of:
    using a write mask as the write command for the write operation.
    [14" id="US-20010003838-A1-CLM-00014] 14. The method according to
    claim 1 , wherein:
    the system is a SCI system with cache coherent non-uniform memory access.
    [15" id="US-20010003838-A1-CLM-00015] 15. A method for performing a write purge partial operation on a memory with a sequential sharing list in a SCI based multi-node system cache coherent non-uniform access, the method comprising the steps of:
    issuing a memory_write_purge_attach_to_list command that attaches a new node to the sharing list; and
    maintaining a connection between the memory and the sharing list while the new node is attaching to the sharing list.
    [16" id="US-20010003838-A1-CLM-00016] 16. The method according to
    claim 15 , wherein:
    the memory is marked fresh prior to the step of issuing a memory_write_purge_attach_to_list command.
    [17" id="US-20010003838-A1-CLM-00017] 17. The method according to
    claim 15 , wherein:
    the sharing list comprises a plurality of nodes, with a first node of the list marked head_fresh, and with a last node of the list marked tail_valid.
    [18" id="US-20010003838-A1-CLM-00018] 18. The method according to
    claim 17 , wherein:
    the remaining nodes of the list are marked mid_valid.
    [19" id="US-20010003838-A1-CLM-00019] 19. The method according to
    claim 15 , further comprising the steps of:
    issuing a cpurge_fresh_to_invalid command from the new node to a first node of the list;
    issuing a cpurge_valid_to_invalid command from the new node to the next node of the list;
    repeating the cpurge_valid_to_invalid command until the node in the list has received the command; and
    issuing a memory update_list-to-home command after the step of repeating has been completed.
    [20" id="US-20010003838-A1-CLM-00020] 20. The method according to
    claim 19 , wherein:
    the memory is marked home subsequent to the step of issuing a memory update_list-to-home command.
    [21" id="US-20010003838-A1-CLM-00021] 21. The method according to
    claim 15 , further comprising the step of:
    using a send request packet to distribute the command for the write operation throughout the system.
    [22" id="US-20010003838-A1-CLM-00022] 22. The method according to
    claim 15 , further comprising the step of:
    using a write mask as the write command for the write operation.
    [23" id="US-20010003838-A1-CLM-00023] 23. A multi-node system that uses a write operation on a memory with a sharing list, the system comprising:
    means for attaching a new node to the sharing list; and
    means for maintaining a connection between the memory and the sharing list while the new node is attaching to the sharing list.
    [24" id="US-20010003838-A1-CLM-00024] 24. The system according to
    claim 23 , wherein:
    the sharing list, the new node, and the memory are connected by a plurality of pointers; and
    the connection is between the new node and the memory.
    [25" id="US-20010003838-A1-CLM-00025] 25. The system according to
    claim 23 , further comprising:
    means for purging each node in the sharing list; and
    means for separating the sharing list from the memory after each node has been purged.
    [26" id="US-20010003838-A1-CLM-00026] 26. The system according to
    claim 25 , wherein:
    the means for purging invalidates the data stored at each node.
    [27" id="US-20010003838-A1-CLM-00027] 27. The system according to
    claim 25 , further comprising:
    means for separating each node in the sharing list from each other.
    [28" id="US-20010003838-A1-CLM-00028] 28. The system according to
    claim 23 , wherein:
    the system is a SCI system with cache coherent non-uniform memory access.
    类似技术:
    公开号 | 公开日 | 专利标题
    US5197146A|1993-03-23|Method for maintaining cache coherence in a multiprocessor computer system
    US5765187A|1998-06-09|Control system for a ring buffer which prevents overrunning and underrunning
    JP5575870B2|2014-08-20|Satisfaction of memory ordering requirements between partial read and non-snoop access
    US4430701A|1984-02-07|Method and apparatus for a hierarchical paging storage system
    US5903911A|1999-05-11|Cache-based computer system employing memory control circuit and method for write allocation and data prefetch
    JP2004348734A|2004-12-09|Method for providing atomic update primitive in asymmetric heterogeneous multiprocessor environment
    US7555610B2|2009-06-30|Cache memory and control method thereof
    KR100505695B1|2005-08-03|Cache memory device having dynamically-allocated or deallocated buffers, digital data processing system comprising it and method thereof
    JP2006277762A|2006-10-12|Divided nondense directory for distributed shared memory multi-processor system
    JP3132749B2|2001-02-05|Multiprocessor data processing system
    US5530833A|1996-06-25|Apparatus and method for updating LRU pointer in a controller for two-way set associative cache
    JPH10232835A|1998-09-02|Sci system increment updating method
    KR20070040340A|2007-04-16|Disable write back on atomic reserved line in a small cache system
    JP2003186743A|2003-07-04|Method and device for locking and unlocking of adaptive cache frame
    US6182164B1|2001-01-30|Minimizing cache overhead by storing data for communications between a peripheral device and a host system into separate locations in memory
    US6381657B2|2002-04-30|Sharing list for multi-node DMA write operations
    JP2004240616A|2004-08-26|Memory controller and memory access control method
    US6256710B1|2001-07-03|Cache management during cache inhibited transactions for increasing cache efficiency
    US6678800B1|2004-01-13|Cache apparatus and control method having writable modified state
    US6496907B1|2002-12-17|System and method for updating from a read-only to a read-write entry and concurrently invalidating stale cache copies from head-to-tail and tail-to-head directions
    US7213109B1|2007-05-01|System and method for providing speculative ownership of cached data based on history tracking
    US6480940B1|2002-11-12|Method of controlling cache memory in multiprocessor system and the multiprocessor system based on detection of predetermined software module
    JP2580263B2|1997-02-12|Buffer storage device
    US6584550B1|2003-06-24|System and method for updating a head entry from read-only to read-write and allowing a list to expand in a cache-coherence sharing list
    US6321304B1|2001-11-20|System and method for deleting read-only head entries in multi-processor computer systems supporting cache coherence with mixed protocols
    同族专利:
    公开号 | 公开日
    JP3745108B2|2006-02-15|
    US6381657B2|2002-04-30|
    JPH10293719A|1998-11-04|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US20050027899A1|2003-07-31|2005-02-03|International Business Machines Corporation|Cacheable DMA|
    US20050216610A1|2004-03-25|2005-09-29|International Business Machines Corporation|Method to provide cache management commands for a DMA controller|JPH0731662B2|1986-07-15|1995-04-10|富士通株式会社|Multiprocessor system|
    US5251303A|1989-01-13|1993-10-05|International Business Machines Corporation|System for DMA block data transfer based on linked control blocks|
    US5297269A|1990-04-26|1994-03-22|Digital Equipment Company|Cache coherency protocol for multi processor computer system|
    JPH0619785A|1992-03-27|1994-01-28|Matsushita Electric Ind Co Ltd|Distributed shared virtual memory and its constitution method|
    JP2566717B2|1992-03-30|1996-12-25|インターナショナル・ビジネス・マシーンズ・コーポレイション|Conditional operation providing apparatus and method|
    US5434976A|1992-09-28|1995-07-18|Standard Microsystems Corporation|Communications controller utilizing an external buffer memory with plural channels between a host and network interface operating independently for transferring packets between protocol layers|
    US5598568A|1993-05-06|1997-01-28|Mercury Computer Systems, Inc.|Multicomputer memory access architecture|
    US5535116A|1993-05-18|1996-07-09|Stanford University|Flat cache-only multi-processor architectures|
    US5446915A|1993-05-25|1995-08-29|Intel Corporation|Parallel processing system virtual connection method and apparatus with protection and flow control|
    US5845310A|1993-12-15|1998-12-01|Hewlett-Packard Co.|System and methods for performing cache latency diagnostics in scalable parallel processing architectures including calculating CPU idle time and counting number of cache misses|
    US5560027A|1993-12-15|1996-09-24|Convex Computer Corporation|Scalable parallel processing systems wherein each hypernode has plural processing modules interconnected by crossbar and each processing module has SCI circuitry for forming multi-dimensional network with other hypernodes|
    US5577204A|1993-12-15|1996-11-19|Convex Computer Corporation|Parallel processing computer system interconnections utilizing unidirectional communication links with separate request and response lines for direct communication or using a crossbar switching device|
    US5875352A|1995-11-03|1999-02-23|Sun Microsystems, Inc.|Method and apparatus for multiple channel direct memory access control|
    US6002757A|1996-03-08|1999-12-14|Stentor Resource Centre, Inc.|Number portability using enhanced routing table|
    US5940600A|1996-04-01|1999-08-17|Apple Computer, Inc.|Isochronous channel having a linked list of buffers|
    US5802578A|1996-06-12|1998-09-01|Sequent Computer Systems, Inc.|Multinode computer system with cache for combined tags|
    US5900020A|1996-06-27|1999-05-04|Sequent Computer Systems, Inc.|Method and apparatus for maintaining an order of write operations by processors in a multiprocessor computer to maintain memory consistency|
    US5887135A|1996-08-15|1999-03-23|International Business Machines Corporation|System and method for management of object transitions in an external storage facility accessed by one or more processors|
    US5717897A|1996-09-09|1998-02-10|Unisys Corporation|System for coordinating coherency of cache memories of multiple host computers of a distributed information system|
    US6041376A|1997-04-24|2000-03-21|Sequent Computer Systems, Inc.|Distributed shared memory system having a first node that prevents other nodes from accessing requested data until a processor on the first node controls the requested data|
    法律状态:
    1997-06-06| AS| Assignment|Owner name: HEWLETT-PACKARD COMPANY, CALIFORNIA Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:HORNUNG, BRYAN;MARIETTA, BRYAN;REEL/FRAME:008546/0299 Effective date: 19970128 |
    2001-01-16| AS| Assignment|Owner name: HEWLETT-PACKARD COMPANY, COLORADO Free format text: MERGER;ASSIGNOR:HEWLETT-PACKARD COMPANY;REEL/FRAME:011523/0469 Effective date: 19980520 |
    2002-04-12| STCF| Information on status: patent grant|Free format text: PATENTED CASE |
    2005-10-31| FPAY| Fee payment|Year of fee payment: 4 |
    2009-10-30| FPAY| Fee payment|Year of fee payment: 8 |
    2011-04-26| AS| Assignment|Owner name: SAMSUNG ELECTRONICS CO., LTD., KOREA, REPUBLIC OF Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:HEWLETT-PACKARD DEVELOPMENT COMPANY, L.P.;HEWLETT-PACKARD COMPANY;REEL/FRAME:026198/0139 Effective date: 20101019 |
    2013-09-24| FPAY| Fee payment|Year of fee payment: 12 |
    优先权:
    申请号 | 申请日 | 专利标题
    US08/792,077|US6381657B2|1997-01-31|1997-01-31|Sharing list for multi-node DMA write operations|US08/792,077| US6381657B2|1997-01-31|1997-01-31|Sharing list for multi-node DMA write operations|
    JP01535998A| JP3745108B2|1997-01-31|1998-01-28|Multi-node system and method for performing write operation of memory with shared list|
    [返回顶部]