Evaluation Methodology for Data Communication-Aware Application Partitioning

Heterogeneous multicore architectures appear to be a promising solution to the ever-increasing computational demands of applications in embedded as well as high-performance computing domains. Nevertheless, inefficient application partitioning for such systems may result in a considerable reduction in the anticipated performance improvement. Consequently, proper mechanisms are required to evaluate quality of different partitions in an early phase of application mapping. This evaluation is especially important in the case of reconfigurable accelerators, as the synthesis phase of their hardware blocks is quite time-consuming. Hence, an early evaluation of application partitioning mechanisms for such architectures can reduce time-to-market. In this paper, we propose a data communication-aware methodology for evaluating the quality of application partitions as well as partitioning algorithms. We also present an open source tool which implements the proposed methodology. Moreover, we evaluate several heuristic algorithms to further substantiate the applicability of the proposed methodology and the utilization of the developed tool. The applications are used as test inputs for the sake of comparisons in terms of relative and absolute quality measurements of the partitioning solutions.