Adaptive cache coherence over a high bandwidth broadband mesh network

Networks have traditionally been an obstacle to high performance distribute d computing. Specific problems are insufficient bandwidth and long transact ion latencies. While pipelining data can achieve high bandwidth, it does no thing for latency which is still a bottleneck in performance. One approach is to develop a cache coherence protocol which exploits recurring data shar ing patterns to reduce the impact of latency. This paper proposes an adapti ve cache coherence protocol which detects producer-consumer type sharing an d maintains coherence on only those cache blocks which exhibit producer-con sumer sharing via updates rather than invalidates. Execution driven simulat ions of this protocol show improved performance compared to a standard writ e-invalidate protocol protocol and a competitive update protocol. When ther e are no access patterns to exploit, the protocol does not degrade performa nce. When there is producer-consumer type sharing, the proposed protocol ru ns benchmarks up to 30% faster than the better of either write-invalidate o r competitive update. As a side-effect, it shows improved tolerance of incr easing network latency. (C) 2000 Published by Elsevier Science B.V. All rig hts reserved.