FAULT-TOLERANT DISTRIBUTED COMPUTING ON NETWORKS OF WORKSTATIONS

A "HYPERCOMPUTER" IS THE PARALLEL COMPUTER THAT EMERGES WHEN ONE SUMS THE UNUSED CYCLES OVER ALL THE COMPUTER NODES ON A LOCAL AREA NETWORK. HOWEVER, FAULT-TOLERANCE IS A CRITICAL ISSUE FOR LONGER-LIVED APPLICATIONS RUNNING ON TYPICAL LANS WHERE NODE FAILURES ARE NOT UNCOMMON. PROCESS CHECKPOINT AND RESTART MECHANISMS THAT ALLOW PARALLEL NETWORK COMPUTATIONS TO PROCEED EVEN AS THEIR CONSTITUENT PROCESSES FAIL ARE BEING DEVELOPED. THE RESULTING SYSTEM IS A FAULT-TOLERANT, EXTENSIBLE, COST-EFFECTIVE SUPERCOMPUTER BASED ON LOW COST, HIGH PERFORMANCE SCIENTIFIC WORKSTATIONS CONNECTED VIA A LOCAL AREA NETWORK.