High performance computing applications run on massively parallel supercomputers consist of concurrent programs designed using multi-threaded, multi-process models. The applications may consist of various constructs (threads, local processes, distributed processes, etc.) with varying degree of parallelism. Although high performance concurrent programs use similar design patterns, models and principles as that of sequential programs, unlike sequential programs, they typically demonstrate non-deterministic behavior. The probability of bugs increases with the number of interactions between the various parallel constructs. Race conditions, data races, deadlocks, missed signals and live lock are common error types.
