Tracking runtime concurrent dependences in java threads using thread control profiling

Abstract

More than 75% of recent Java projects include some form of concurrent programming. Due to complex interactions between multi-threads, concurrent programs are often harder to understand and test than single threaded programs. To facilitate understanding and testing of concurrent programs, we developed a new profiling method called TCP (Thread Control Profiling). Outputs of TCP presents frequencies of control dependence, which includes thread creation, thread synchronization, interruption, and so on, of the executed thread. TCP first performs static analysis of detailed concurrency syntax and semantics of Java to construct the profiling graph model TCDG (Thread Control Dependence Graph). TCDG is then used for instrumentation and for generating profiles. We have evaluated TCP using a case study and a few experiments. The case study shows that TCP method can effectively prioritize test cases for testing concurrent programs. One experiment shows that outputs from TCP facilitate developers’ understanding of concurrent code. Other experiments evaluate various possible overheads introduced by the TCP method. Results show that TCP can provide rich and useful information with reasonable costs.