Icare: Combining Efficiency and High-availability in a Dsm System

In light of the increasing throughput of local area networks, Networks Of Workstations (NOW) which provide a distributed shared memory (DSM) have become a convenient alternative to parallel architectures in the framework of parallel scientific applications. ICARE is a recoverable DSM based on backward error recovery which is implemented on top of an experiments ATM platform running the CHORUS microkernel. ICARE provides a way to combine both efficiency and high availability in such a system. The fact that recovery data are stored in standard memory provides a low-cost fault-tolerance mechanism as well as the opportunity to exploit the symbiotic relationship between the data replication existing in essence in DSM systems and the one needed for fault-tolerance to increasing efficiency. Furthermore, ICARE implements an efficient transparent rollback recovery. Indeed, the disturbance of a rollback is minimized by distributing the load in a balanced way among the remaining fault-free nodes in the event of a permanent failure. Performance evaluations show the efficiency of the ICARE prototype that implements the proposed algorithms. Key-words: Network of Workstations, Distributed Shared Memory, Backward Error Recovery, Data Replication, Efficiency