Developing AWorkflow Management System for Teamster-G

Abstract

We have successfully developed an open source computational grid system called Teamster-G with the support of National Science Councils. This system enables users to develop parallel grid applications by means of the shared memory interface such as Pthread and OpenMP. Moreover, it provides users with transparent resource allocation and reconfiguration services to optimize the execution performance of user jobs. However, Teamster-G currently allows users to submit one job at one time. Therefore, it is difficult and inconvenient for users to develop workflow applications such as E-science and Grid ERP on Teamster-G. To address this problem, we develop a workflow management system for Teamster-G in this study. This system provides users with a set of web-services including workflow edition, submission, execution, and monitoring. Moreover, we proposed a new resource-selection algorithm based on long-term resource-availability prediction. Our experimental results have shown that the proposed algorithm is effective for reducing the makespans of workflows.