Heterogeneous Distributed Shared Memory

Heterogeneity in distributed systems is increasingly a fact of life, due to specialization of computing equipment. It is highly desirable to integrate heterogeneous hosts into a coherent computing environment to support distributed and parallel applications, so that the individual strengths of the different hosts can be exploited together. Distributed shared memory (DSM), a high-level, highly transparent model for interprocess communication in distributed systems, is a promising vehicle for achieving such an integration. This paper studies the design, implementation, and performance of heterogeneous distributed shared memory (HDSM). As a practical research effort, we have developed a prototype HDSM system that integrates very different types of hosts, and have ported a number of applications to this system. Our experience shows that, despite a number of difficulties in data conversion, HDSM is indeed implementable with minimal loss in functional and performance transparency when compared to homogeneous DSM systems.