High-performance mixed-machine heterogeneous computing

The focus of this invited keynote paper (to represented by H. J. Siegel) is mixed-machine heterogeneous computing, where a suite of different kinds of high-performance machines are interconnected by high-speed links. Such a system can be orchestrated to perform an application whose subtaslw have diverse execution requirements. $ubtasks are assigned to and executed on the machines that will result in a minimal overall execution time for the task, considering factors including inter-machine communication overhead. A conceptual model of the automatic decomposition of tasks and assignment of subtasks is presented. Examples of static and dynamic approaches to the matching and scheduling of subtasks are summarized. Some open research problems are dwcussed.