A comparative study of DAG clustering

Organizing tasks that are decomposed from workflows with directed acyclic graph (DAG) is a common practice. Assigning the tasks in DAG to physical computing nodes is a critical step for minimizing the total workflow processing time. However, scale and diversity of the DAG increase distinctly as the increment of the complexity of applications. Waiting time introduced by the dependencies between tasks affect the processing time of workflows severely. Cluster based task assignment is promising for reducing the waiting time introduced by dependencies. In which the key element is the cluster method that are taken to group the tasks. This paper comparatively studied the task assignment performance with different DAG clustering methods. The experiment results show that genetic based clustering method is better in reducing the make-span and enlarging the speedup for workflows.