Evaluating The Performance Of Four Snooping Cache Coherency Protocols

摘要: Write-invalidate and write-broadcast coherency protocols have been criticized for being unable to achieve good bus performance across all cache configurations. In particular, write-invalidate can suffer as block size increases; large sizes will hurt write-broadcast. Read-broadcast competitive snooping extensions the proposed solve each problem. Our results indicate that benefits of are limited. reduces number invalidation misses, but at a high cost in processor lockout from cache. The net effect be an increase total execution cycles. Competitive only those programs with per-processor locality Of reference shared data. For characterized by inter-processor contention addresses, competi- tive degrade causing slight utilization time.