Hybrid Balanced Task Clustering Algorithm for Scientific Workflows in Cloud Computing

Abstract

Scientific Workflow is a composition of both coarse-grained and fine-grained computational tasks displaying varying execution requirements. Large-scale data transfer is involved in scientific workflows, so efficient techniques are required to reduce the makespan of the workflow. Task clustering is an efficient technique used in such a scenario that involves combining multiple tasks with shorter execution time into a single cluster to be executed on a resource. This leads to a reduction of scheduling overheads in scientific workflows and thus improvement of performance. However available task clustering methods involve clustering the tasks horizontally without the consideration of the structure of tasks in a workflow. We propose hybrid balanced task clustering algorithm that uses the parameter of impact factor of workflows along with the structure of workflow. According to this technique, tasks can be considered for clustering either vertically or horizontally based on the value of the impact factor. This minimizes the system overheads and the makespan for execution of a workflow. A simulation based evaluation is performed on real workflows that shows the proposed algorithm is efficient in recommending clusters. It shows improvement of 5-10\% in makespan time of workflow depending on the type of workflow used.