Hybrid CPU and GPGPU Volunteer Computing Framework over the Extensible Messaging and Presence Protocol for Prallel Branch and Bound Optimization of Truss Structures
Poznan Supercomputing and Networking Center, ul. Noskowskiego 10, 61-704, Poznan, Poland
HeteroPar, 2011
@article{mackowiak2011hybrid,
title={Hybrid CPU and GPGPU Volunteer Computing Framework over the Extensible Messaging and Presence Protocol for Prallel Branch and Bound Optimization of Truss Structures},
author={Mackowiak, M. and Kurowski, K. and Igumenov, A. and Zilinskas, J.},
year={2011}
}
A new fault-tolerant message passing interface framework for a hybrid parallel com-puting in distributed volunteer-based networks is presented in this paper. It is a new programming core library being able to link together many CPUs and promis-ing General Purpose Graphic Processing Unit (GPGPU) devices over the Internet. We successfully developed a high-level programming interface on the top of well-known XMPP communication protocol to exchange and synchronize controlling messages among computing processes executed remotely via the Internet on dis-tributed computing clusters. In this short paper, we briefly present the architecture of our fault-tolerant parallel programming framework called XMPP-MPI, together with preliminary results obtained. We conclude this paper with best practices and improvements that will be needed to solve computational problems like presented parallel branch an bound algorithm for topology optimization of truss structures using larger amount of volunteer computing resources.
December 27, 2011 by hgpu
