在多緒執行系統中，是否適當分配所擁有的資源將對程式的效率造成很大的影響。在一個分散式環境中的程式將會有更顯著的影響，這導因於資源在此一環境中的更變性較大；軟體分散共享記憶體系統正是一個需要良好資源配置以達高效率運算的環境。這篇論文提出一個以機器與記憶體分頁之間的關係為主要考量的軟體分散式共享記憶體系統；在此系統中，我們將根據各執行緒的記憶體取用模式來分配記憶體分頁，並視情況讓效率較差的執行緒得以收集需要的分頁到該台機器上，並藉此減少記憶體分頁錯誤的發生次數來提升效率。然而，考慮到程式平行度，我們應避免讓所有的記憶體分頁都被收集到同一機器上來提升記憶體局部性；因此，我們加入了執行緒遷移的機制來適應於分頁分散於各台機器的情況。另外，我們採用InfiniBand作為網路傳輸介面；藉著使用InfiniBand的各種特性，如遠端直接記憶體存取(RDMA)、分散與聚集功能等，效率將能夠更進一步的被提升。實驗結果顯示我們的分散式共享記憶體系統MLDSM對於記憶體存取顆粒度大的程式能夠有良好的表現，而同時對於程式存取顆粒度較小的情況，也可避免其所造成的大量負擔。

In a multithreaded system, whether the arrangement of resources properly or not can make a big difference on efficiency. It is even more remarkable for a distributed one, because the resources are more dynamic. A software DSM system is such an environment that needs good management to achieve high performance. This thesis presents a software DSM system that concerns about the relationship between memory pages and nodes. In this system, we try to arrange pages according to the access pattern of threads. When a thread suffers from high protocol overhead, pages will be aggregated by the thread on the same node to improve performance by reducing the times of page fault happened. However, for parallelism consideration, we should avoid grouping all pages on the same node to enhance the locality. Hence we add thread migration mechanism into the system while the pages are still distributed among nodes. We also adopted InfiniBand as the communication network. By using the features of InfiniBand, such as remote direct memory access (RDMA), scatter-gather functionality, etc., the performance can be improved further. The results show that our software DSM system, MLDSM, performs well in programs with coarse-grain access patterns and avoid the situation with high protocol overheads due to finer-grain access patterns. The MLDSM can be adaptive to the behaviors of programs.

[1] G. Antoniu, L. Bougé, DSM-PM2: A portable implementation platform for multithreaded DSM consistency protocols. In Proceedings of the 6th International Workshop on High-Level Parallel Programming Models and Supportive Environments (HIPS '01), pp. 55-70, 2001.
[2] G. Antoniu, C. Perez, Using Preemptive Thread Migration to Load-balance Data-parallel Applications. In Proceedings of the 5th International European Conference on Parallel Processing (Euro-Par '99), pp. 117-124, 1999.
[3] K. Gharachorloo, D. Lenoski, J. Laudon, P. Gibbons, A. Gupta, and J. Hennessy, Memory Consistency and Event Ordering in Scalable Schared-Memory Multiprocessors. In Proceedings of the 17th International Symposium on Computer Architecture (ISCA '90), pp. 15-26, May 1990.
[4] L. Iftode, Home-based Shared Virtual Memory. Ph. D. thesis, Princeton University, Aug 1998.
[5] S. Jenks, J. L. Gaudiot, Nomadic Threads: A Migrating Multithreaded Approach to Remote Memory Accesses in Multiprocessors. In Proceedings of the International Conference on Parallel Architectures and Compilation Techniques (PACT '96), pp. 2-11, October 1996.
[6] P. J. Keleher, A. L. Cox, and W. Zweanepoel, Lazy Release Consistency for Software Distributed Shared Memory. In Proceedings of the 19th International Symposium of Computer Architecture (ISCA '92), pp. 13-21, May 1992.
[7] K. Li, IVY: A Shared Virtual Memory System for Parallel Computing. In Proceedings of the 1988 International Conference on Parallel Processing, vol. 2, pp. 94-101, August 1988.
[8] Chang-Yi Lin, An Adaptive Migratory Home Protocol for Software DSM Systems. Master thesis, Tsing-Hua University, Jun 2004.
[9] L. Liss, Y. Birk, A. Schuster, Efficient Exploitation of Kernel Access to InfiniBand: a Software DSM Example. In Proceedings of the 11th Symposium on High Performance Interconnects (HOTI '03), pp. 130-135, August 2003.
[10] W. Y. Liang, C. T. King, and F. Lai, Adsmith: An Object-based Distributed Shared Memory for Network of Workstations. In IEICE Transaction on Information and Systems, vol. E80-D, no. 9, pp. 899-908, Sep 1997.
[11] Y. T. Liu, T. Y. Liang, C. T. Huang, C. K. Shieh, Memory Resource Considerations in the Load Balancing of Software DSM Systems. In Proceedings of the 2003 International Conference on Parallel Processing Workshops (ICPPW '03), pp. 71-78, 2003.
[12] D. S. Nikolopoulos, C. D. Polychronopoulos, Scheduler-Activated Dynamic Page Migration for Multiprogrammed DSM Multiprocessors. In Journal of Parallel and Distributed Computing, vol. 62, pp. 1069-1103, June 2002.
[13] R. Noronha, D. K. Panda, Designing High Performance DSM Systems using InfiniBand Features. In Proceedings of the 4th International Symposium on Cluster Computing and the Grid (CCGrid '04), pp. 467-474, April 2004.
[14] C. Osendorfer, J. Tao, C. Trinitis, M. Mairandres, ViSMI: Software Distributed Shared Memory for InfiniBand Clusters. In Proceedings of the 3rd International Symposium on Network Computing an Applications (NCA '04), pp. 185-191, August 2004.
[15] A. Rogers, M. C. Carlisle, J. H. Reppy, L. J. Hendren, Supporting Dynamic Data Structures on Distributed Memory Machines. In ACM Transactions on Programming Lanaguages and Systems, vol. 17, no. 2. pp. 233-263, 1995.
[16] K. Thitikamol, P. J. Keleher, Multi-threading and Remote Latency in Software DSMs. In Proceedings of the 17th International Conference on Distributed Computing System (ICDCS '97), pp. 296-304, 1997.
[17] K. Thitikamol, P. J. Keleher, Thread Migration and Communication Minimization in DSM Systems. In Proceedings of the IEEE, Special Issue on Distributed Shared Memory, pp. 487-497, 1999.
[18] L. Whately, R. Pinto, R. Bianchini, C. Amorim, Adaptive Techniques for Home-Based Software DSMs. In Proceedings of the 13th Symposium on Computer Architecture and High-Performance Computing, September 2001.
[19] Y. Zhou, L. Iftode, and K. Li, Performance Evaluation of Two Home-Based Lazy Release Consistency Protocols for Shared Virtual Memory Systems. In Proceeding of the 2nd Symposium on Operating System Design and Implementation (OSDI '96), pp. 75-88, October 1996.