An adaptive scheduling approach in distributed systems

A large number of scheduling algorithms for large scale distributed systems have been proposed, studied and compared. In spite of this, there are few studies comparing the performance of scheduling algorithms considering at the same time the distributed system structure on which we want to schedule the tasks; the type of directed acyclic graph (DAG), in which graph nodes represent tasks and graph edges represent data transfers; and the type of tasks, for example CPU-bound vs. I/O bound. This paper proposes a method for selecting, in a dynamic manner, the most appropriate scheduling algorithm for a particular distributed system, after a previous analysis. Choosing the best known scheduling algorithm can improve performance of an application if all the aspects previously enumerated are considered. The results of this paper consist in a comparison of the scheduling algorithms performance for the following scheduling algorithms: Modified Critical Path, Cluster ready Children First, Earliest Time First, Highest Level First with Estimated Times and Hybrid Remapper Minimum Partial Completion Time Static Priority. However, the main purpose of this investigation tests is to demonstrate the usefulness of the presented scheduling approach using these results.