Silkroad : A system supporting DSM and multiple paradigms in cluster computing

Abstract

In Cluster Computing, programming paradigms is an important high-level issue that defines the way to structure algorithms to run on a parallel system. However, Few parallel systems support multiple parallel programming paradigms. To address this problem, we developed a graph theory to analyze programming paradigms and explore multiple paradigms with the memory model approach. We show that by strengthening the memory consistency model of shared virtual memory based systems, the range of the supported paradigms can be widened. This approach is implemented in our SilkRoad runtime system, which is a variant of Cilk system. It shows that with our strengthened consistency model (RC_dag consistency), SilkRoad can efficiently run applications inherited from Cilk while support wider paradigms (Divide and Conquer, SPMD, etc). Our approach sheds light on achieving multiple paradigms both theoretically and empirically.