Is enhancing the functionality of process grouping abstractions in parallel programming libraries desirable and feasible?

Recent research has shown an increasing interest in accommodating within a single environment two forms of exploitation of parallelism. This includes parallelism within a single homogeneous machine and parallelism across a network of possibly heterogeneous machines. From our point of view, current programming systems for this kind of environment lack structuring abstractions for cooperative computing. In fact, although support for process grouping is provided, it does not address issues of distinct event orderings such, as uniform message delivery and causality (in the "happens before" sense of Lamport, 1978), which are fundamental for the preservation of consistent views among the involved cooperating processes. We argue that support for causality and consistency in the process views of parallel computations, as related to process grouping abstractions, are fundamental devices for understanding and building concurrent systems with asynchronous components with distributed memory (and local states). A better user understanding of parallel computations concerning their correctness, dynamic behavior, and performance, needs specific support from the system regarding such aspects. The above ideas are being integrated in a virtual machine environment called DVM-that we are designing and implementing.<
>