発表論文

題名 Pruning Cache: A Dynamic Directory Generation Scheme For Distributed Shared Memory

著者西村克信，工藤知宏，西宏章，天野英晴

概要 The pruning cache directory method is proposed for dynamic hierarchical directory management on large scale CC-NUMA systems in which a node aggressively shares data with other nodes. By a combination with Reduced Hierarchical Bitmap Directory (RHBD) method, dynamic directory is quickly formed and managed with a small (about 5%) additional hardware.
The result of a trace driven simulation of a system with 64 processors shows that more than 90% hit ratio can be obtained with a small 2-way associative pruning cache with 32 entries. From the probabilistic simulation of a large system, it appears that the combination of the pruning cache and the RHBD achieves better latency and bandwidth than those of traditional directory management methods based on 1-to-1 message passing.

発表 Proceedings of the Second IASTED International Conference
Parallel and Distributed Computing and Networks (PDCN'98)　89～94頁
1998年12月

論文はこちらです（英文、PostScript + gzip 形式、302KB）

題名	Pruning Cache: A Dynamic Directory Generation Scheme For Distributed Shared Memory
著者	西村克信，工藤知宏，西宏章，天野英晴
概要	The pruning cache directory method is proposed for dynamic hierarchical directory management on large scale CC-NUMA systems in which a node aggressively shares data with other nodes. By a combination with Reduced Hierarchical Bitmap Directory (RHBD) method, dynamic directory is quickly formed and managed with a small (about 5%) additional hardware. The result of a trace driven simulation of a system with 64 processors shows that more than 90% hit ratio can be obtained with a small 2-way associative pruning cache with 32 entries. From the probabilistic simulation of a large system, it appears that the combination of the pruning cache and the RHBD achieves better latency and bandwidth than those of traditional directory management methods based on 1-to-1 message passing.
発表	Proceedings of the Second IASTED International Conference Parallel and Distributed Computing and Networks (PDCN'98)　89～94頁 1998年12月