Local State Reusing for Efficient Model Checking of Multithreaded Programs

Abstract

Applying model checking to detect concurrency errors in larger-scale multithreaded programs is limited by state explosion problem stemming from nondeterminism. We propose a novel approach established on the insight into the relationship between thread interference and nondeterminism to break the limitation. The approach works for particular parallel region that can be divided into disjoint groups among which there is no thread interference. We demonstrate that the set of reachable states of the parallel region is the Cartesian product of reachable states of each disjoint group. Local states of disjoint groups explored in previous runs can be reused to avoid redundant state transitions such that the time consumed by successive runs is decreased. The empirical results indicate that the efficiency of model checking can be improved by orders of magnitude through local state reusing.