Fast Generation of Large Task Network Mappings

Abstract

In the context of networks of massively parallel execution models, optimizing the locality if inter-process communication is a major performance issue. We propose two heuristics to solve a dataflow process network mapping problem, where a network of communicating tasks is placed into a set of processors with limited resource capacities, while minimizing the overall communication bandwidth between processors. Those approaches are designed to tackle instances of over one hundred thousand tasks in acceptable time.