Load balanced parallel program verification on a shared memory machine

Writing parallel programs that encompass explicit parallelism is still a difficult task. Appropriate development tools are needed. This paper describes the design and the parallel implementation of a verification tool based on static analysis. It supports compile-time deadlock detection of CSP-based programming systems (e.g. Occam2). AND/OR-parallelism is used to parallelise the algorithm, which is basically a search problem. To support dynamic load balancing in an efficient way, a 'Teamsystem' is designed, which supports the management and migration of fine-grained tasks. A hybrid load balancing protocol is proposed This means that both a receiver- and a sender-initiated strategy is employed to redistribute the work at run-time. The parallel implementation on a shared memory machine (Sequent Symmetry) resulted in a linear speedup.<
>