![]() ![]() ![]() |
![]() |
|
|
![]() ![]() ![]() ![]() ![]() |
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Return to Implementation Issues In a previous paper we proposed a disk cache coherence algorithm, called DSMIO, for multicomputers or clusters of workstations. The algorithm keeps caches coherent using concepts and mechanisms from software distributed shared-memory systems, such as multiple concurrent writers and the Lazy Release Consistency model for reducing the negative impact of false sharing. In this paper we present a detailed analysis of the performance of DSMIO in the context of a parallel object-based database system where one node acts as a server, accessing stable storage and keeping caches coherent. We perform experiments with the 007 Benchmarks on an SP2 multicomputer. The most interesting part of the paper compares the DSMIO results against those of the state-of-the-art CallBack Locking (CBL) cache coherence algorithm. The results of this comparison show that our system consistently and significantly outperforms CBL. DiSC'01 Copyright ©2002 ACM Inc. |