A checkpointing algorithm for an SCI based distributed shared memory system

摘要

Distributed Shared Memory (DSM) systems combine the ease of programming of shared memory parallel computers and scalability of meassage passing multicomputers. IEEE has proposed an interface standard known as SCI standard to construct DSM systems. When the number of processors in a parallel computer increase it is imperative to build fault tolerance. This article presents an algorithm for checkpointing and rollback recovery of an SCI based DSM system using the provisions of the standard. It is shown that this checkpointing and rollback recovery procedure judiciously combines the features of both shared memory and message passing distributed memory system.