Non-uniform and dynamic domain decompositions for hypercomputing

Implementing efficient parallel programs on a network-based computing platform (hypercomputing) is still a challenge. This paper proposes a new Adaptive Data Distribution (ADD) support that avoids to the programmer the complex task of managing irregular data distributions and adapting them to the nonuniform and variable conditions of a shared platform. In particular, ADD provides a data partition that fits the heterogeneity of the platform at loadtime, a set of data inquiry primitives that allow the programmer to deal with a logical data partition, and a runtime support that during computation adapts the distribution to the variations in platform's available performance. We demonstrate that ADD can autonomously maintain the efficiency of SPMD data parallel computations even on heterogeneous and highly variable computing platforms.