On the robustness of the soft state for task scheduling in large-scale distributed computing environment

In this paper, we consider task scheduling in distributed computing. In distributed computing, it is possible that tasks fail, and it is difficult to get accurate information about hosts and tasks. WQR (workqueue with replication), which was proposed by Cirne et al., is a good algorithm because it achieves a short job-completion time without requiring any information about hosts and tasks. However, in order to use WQR for distributed computing, we need to resolve some issues on task failure detection and task cancellation. For this purpose, we examine two approaches-the conventional task timeout method and the soft state method. Simulation results showed that the soft state method is more robust than the task timeout method.