Many applications make use of hybrid programming models intermixing task and data parallelism in order to exploit modern architectures more efficiently. However, unbalanced computational load or idle times due to tasks that are blocked either in I/O or waiting on results from other tasks can cause significant performance problems. Fortunately, such idle times can be overlapped with useful computation in many cases. In this paper we propose a simple, yet powerful methodology for specifying intra-object parallelism and synchronization in the context of the coordination language Opus. Our design combines both, static and dynamic synchronization on a high level. We motivate our design with some examples and discuss implementation strategies for compilation as well as runtime support.

Top