Petri net based algorithm modelization and parallel execution on symmetric multiprocessors

Abstract

The Symmetric Multiprocessors architecture is composed by a complex set of cores, chips and memory channels that make it difficult to implement a parallel program that efficiently uses all resources. Another obstacle for achieving a performance according the resources is added by algorithms with hard data dependency. Asynchronicity is a key to get all processors running. Petri Nets have been used for a long time to model algorithms, but not as a tool to parallel execution. In this paper we introduce an asynchronous Parallel Execution Model based on Petri Nets and the process to go from a high level model to an executable parallel program. The Cholesky Factorization algorithm is used as a testbed. Tests results yield values that are near the theoretical peak and open good prospects to expand the model to other environments and algorithms.