Web services for MPI-based parallel applications on a rocks cluster

Abstract

MPI-based parallel applications including scientific applications are now widely executed on clusters and grids, and great benefits have been brought to scientific community. However, writing parallel applications would not be easy even for experienced programmers. In this paper, we propose the design and implementation of the MPI-SV middleware that connects gSOAP, the web service interface, and MPICH, the parallel software tool running on a Rocks cluster. With MPI-SV, users can automatically call parallel functions executed on a cluster as if they were regular functions. Hence, users do not have to write MPI-based parallel functions themselves. MPI-SV middleware is implemented on a Rocks cluster and complies with the SOAP standard specification so that it could interoperable with other web services. Three experiments based on numerical calculations are conducted, and the results show that the response time is increased when the data size increases. Even though sending data in MPI-SV via XML in SOAP incurs high communication overhead, our experimental results show that, for a parallel application requiring high computation, the overhead time would be low when compared with computation time. Thus, MPI-SV would give good performance for applications that the computation time dominates the communication time, and fortunately most scientific applications have high computation. ©2008 IEEE.