The impact of caching in a loosely-coupled clustered software DSM system

As interconnected local-area workstation networks are widely available, the idea of offering a software distributed shared memory (SDSM) layer across them is quite an attractive alternative for compute-intensive applications. However, the higher cost of sending a message over an inter-cluster link than over an intracluster can limit applications' performance on a multicluster SDSM system. In this paper, we present the extensions that we have added to TreadMarks SDSM in order to adapt it to a loosely-coupled cluster-based platform. We have implemented a logical per-cluster cache in order to exploit cluster locality. By accessing its local cache, a processor can share data previously requested by another processor of its cluster, thereby hiding the cost of inter-cluster communication.