Improving both fault tolerance and performance by replicating control data in scope consistent software DSMs

To use the shared memory programming paradigm in distributed architectures where there is no physically shared memory, an abstraction must be created. This abstraction is known as Distributed Shared Memory (DSM). Scope consistent software DSMs provide a relaxed memory model that guarantees consistency only at synchronization operations, in a per-lock basis. As the main goal of DSM systems is to provide support for long term computation intensive applications, a fault tolerance support is highly desirable. This article presents and evaluates a mechanism that replicates scope consistent DSM control data in remote nodes, aiming to provide fault tolerance support. Our results on some popular benchmarks show that the overhead introduced by the proposed mechanism is low and, surprisingly, in some cases, applications ran faster with the replication mechanism than without it, since replicas created by the mechanism are guaranteed to be up-to-date and can be used by the application, reducing the internode communication overhead.