A workflow is a collection of workflow tasks interconnected by workflow control structures according to the workflow process logic. Because the placement of workflow tasks and resources are distributed to several hosts in distributed workflow environments, each workflow task completes its role by accessing system resources located in any hosts and then transfers the work-flow execution control to its neighbor workflow tasks which may be located in remote hosts. If we can assign adjacent workflow tasks as close as possible and keep the workflow tasks close to their required resources, the overhead of workflow processing can be reduced significantly. In this regard, efficient distribution of workflow tasks may be considered as one of the most influential factors on high performance workflow processing. However, this has not been much addressed in the literature so far.
In this thesis, we propose an efficient workflow task allocation method called GM-WTA which is based on the multilevel graph partitioning. This method can improve the performance of workflow processing by minimizing the remote communication costs occurred during workflow execution. The various experimental results show its efficiency compared to the previous methods.