ON CC-NUMA COMPUTE SERVERS

The dominant architecture for the next generation of shared-memory mul tiprocessors is CC-NUMA (cache-coherent nonuniform memory architecture ). These machines are attractive as compute servers because they provi de transparent access to local and remote memory. However the access l atency to remote memory is 3 to 5 times the latency to local memory. C C-NOW machines provide the benefits of cache coherence to networks of workstations, at the cost of even higher remote access latency. Given the large remote access latencies of these architectures data locality is potentially the most important performance issue. Using realistic workloads, we study the performance improvements provided by OS suppor ted dynamic page migration and replication. Analyzing our kernel-based implementation, we provide a detailed breakdown of the costs. We show that sampling of cache misses can be used to reduce cost without comp romising performance, and that TLB misses may not be a consistent appr oximation for cache misses. Finally, our experiments show that dynamic page migration and replication can substantially increase application performance, as much as 30%, and reduce contention for resources in t he NUMA memory system.