Experimental evaluation of objective functions for well-balanced mapping

High performance of parallel computing on a message-passing multicomputer system relies on the balance of the workloads located on the processing elements of the system and the minimum communication overheads among them. Mapping is the technology to partition the problem domain wellbalanced into multiple distinet execution tasks based on some measures. In mapping, a good objective function is the criterion to guarantee the distinct execution tasks equitable. In this paper, we evaluate five categories of those existed objective lanetions with three different problem subjects using experiments and find an objective function is much suitable for all kinds of problems.