The Influence of Efficient Message Passing Mechanisms on High Performance Distributed Scientific Computing

• Published in: 2008 IEEE International Symposium on Parallel and Distributed Processing with Applications pp. 663 - 668
• Main Authors: Mirtaheri, S.L., Khaneghah, E.M., Sharifi, M., Azgomi, M.A.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-12-2008
• Subjects: Computational complexity; Computational modeling; Computer simulation; Delay; Distributed computing; Distributed Inter Process Communication; Distributed Systems; High Performance Scientific Computing; Libraries; Message passing; Molecular Dynamics; Operating System Kernel; Parallel programming; Safety; Scientific computing
• ISBN: 9780769534718; 0769534716
• ISSN: 2158-9178
• DOI: 10.1109/ISPA.2008.131

Parallel programming and distributed programming are two solutions for scientific applications to provide high performance and fast response time in parallel systems and distributed systems. Parallel and distributed systems must provide inter process communication (IPC) mechanisms like message passing mechanism as underlying platforms to enable communication between local and especially geographically dispersed and physically distributed processes. Communication overhead is the major problem in these systems and there are a lot of efforts to develop more efficient message passing mechanisms or to improve the network communication speed. This paper provides hard evidence that an efficient implementation of message passing mechanism on multi-computers reduces the execution time of a molecular dynamics code. A well-known program for macromolecular dynamics and mechanics called CHARMm is executed on a networked cluster. The performance of CHARMm is measured with two distributed implementations of message passing, namely a kernel-level implementation called DIPC2006 and a renowned library level implementation called MPI. It is shown that the performance of CHARMm on a DIPC2006 configured cluster is by far better than its performance on an optimized MPI configured similar cluster. Even ignoring the favorable points of kernel-level implementations, like safety, privilege, reliability, and primitiveness, the insight is twofold. Scientists are nowadays faced with more computational complexity and look for more efficient systems and mechanisms. Efficient distributed IPC mechanisms have direct effect on running scientistspsila simulations faster, and computer engineers may try harder to develop more efficient distributed implementations of IPC.