A System for Load Balancing a Heterogeneous Cluster

R. Chun and K.T.-K. Ho (USA)


Cluster Computing, Load Balancing, Parallel Processing, Heterogeneous Cluster, Scheduling


A load balancing methodology for a parallel processing cluster system consisting of heterogeneous compute nodes is discussed. The method factors into account the suitability of each node’s architectural configuration to process a particular class of application as a function of its instantaneous load. It can also provide estimated time-to completion for processes running on each of the different nodes. Experimental results are reported using a matrix multiplication benchmark program that was run on a small heterogeneous cluster. The development and implementation of the experimental framework, which was built upon two existing open source systems (Chameleon Message Passing Interface and Open PBS), is also described.

Important Links:

Go Back