Debugging concurrent programs

Concurrent programming is much harder than sequential programming. There are multiple reasons for this. First, the details of the memory model are much more important in concurrent programming, resulting in increased programming complexity. Even on a platform with a well-defined memory model, such as the JVM, the programmer must take care to use proper memory access primitives in order to avoid data races. Then, it is harder to track the execution of a multithreaded program, simply because there are multiple executions proceeding simultaneously. Language debuggers are still focused on tracking the execution of a single thread at a time. Deadlocks and inherent nondeterminism are another source of bugs, neither of which is common in sequential programs. To make things worse, all these issues only have to do with ensuring the correctness of a concurrent program. Ensuring improved throughput and performance opens a separate set of problems, and is often harder...