Techniques for Reducing the Impact of Inclusion in Shared Network Cache Multiprocessors Techniques for Reducing the Impact of Inclusion in Shared Network Cache Multiprocessors

This paper investigates design alternatives for shared network caches in clusterbased multiprocessors. Using simulation, we rst demonstrate that network caches o er several potential performance bene ts, but that the adverse impact of cache inclusion-related evictions must be mitigated for these bene ts to be fully realized. We then evaluate three network cache architectural alternatives designed to address this issue: increasing network cache associativity, adding a network victim cache, and adding a tag cache to relax inclusion requirements for clean network cache lines. Six well-known parallel scienti c applications were used in our experiments. We found that a four-way set associative network cache, or a four-entry victim cache, dramatically reduced execution time for all applications examined. Employing a tag cache for replaced clean lines improved performance for some applications, but this bene t was highly dependent upon processor cache associativity.