On Verification of Parallel Message-Passing Processes

Abstract.

One way to reason about parallel processes is to assume that the execution of each process is subdivided into ‘small enough’ steps, and that these are executed in an interleaved fashion, thus obtaining a sequential program. The steps should be so small that for any parallel execution there will, in a suitable sense, exist a corresponding interleaved execution ending in the same state. The usual way to ensure this is to require that each step should contain at most one global access. However, if the global entities are communication channels, then larger steps may in some cases be allowed, and this may make reasoning about the programs easier. This paper explores these cases, and discusses consequences or verification and deadlock avoidance.