Replicating statement execution for fault detection on distributed memory multiprocessors

A compiler-assisted methodology is proposed for fault detection on distributed-memory systems. Selected instances of program statements are replicated in a way that ensures appropriate coverage. Replication strategies for the detection of permanent and transient faults are presented. These strategies use idle processor times for replicating statement execution whenever possible. Two approaches are also discussed for implementing the proposed strategies on single-program multiple-data parallel execution platforms. The first approach replicates program statements through source-to-source program transformations while the second approach achieves the replication of program statements indirectly by replicating data on multiple processors