Ismat: a virtual machine for compositional parallelism

Parallel programming is hard. Developers must not only solve the problem at hand, but do so in a manner which can best utilize the available resources. In this paper we introduce Ismat, a virtual machine which supports parallelism concerns in an abstracted and inherent manner. Ismat endeavors to simplify the transition from a sequential to parallel environment by providing fundamentally sequential semantics in a parallel realm. Developers incrementally introduce parallelism and performance in contrast to parallelism and correctness. Additionally, Ismat provides and promotes parallel composition to solve problems; reducing developer effort and promoting reuse. We present a detailed explanation of the proposed virtual machine including constructs and semantics. We argue that the resulting virtual machine reduces programmer complexity, provides better abstractions and allows for the exploitation of resources on heterogeneous machines for a range of parallel problems.