Reduction Techniques for Distributed Applications

In this section, we explain two different reduction techniques to reduce the state space of distributed multithreaded applications. These techniques are exploited in our work. One technique (Section 1) is a slight modification of a reduction technique presented by Godefroid in [1, Section 2]. This reduction technique by Godefroid reduces the state space of a system consisting of a single process that has multiple threads. These threads are assumed to be deterministic. We apply a similar technique to a system with multiple processes (Section 2). We also show that applying such a technique on the state space of a distributed system preserves deadlocks and assertion violations. Moreover, we propose a partial order reduction technique for distributed systems (Section 3). Our technique relies on the fact that each process has its own local data which is not shared with any other processes in the system. We show that applying our POR algorithm on a system, which is constructed by the former reduction technique, allows for detecting deadlocks. We use the persistent set technique of Godefroid [2] to prove the correctness of our algorithm. Using this technique, we show that in each state, our algorithm explores a sufficient subset of all possible transitions leading out of the state, preserving deadlocks. The reduction techniques are presented in general settings. In the last section (Section ??), we specialize these techniques to JPF. We show that JPF applies the reduction technique of Godefroid [1] for model checking single process Java applications which preserves deadlocks and assertion violations. Moreover, we show that applying our POR algorithm within JPF allows for detecting global deadlocks in distributed Java applications.