Dynamic Load Balancing of Iterative Data Parallel Problems on a Workstation Clustering

Dynamic load balancing on the workloads of the clustered workstations has emerged as a powerful solution for overcoming load imbalance. In order to detect such imbalances, some load balancing methods check the average idle-time of the workstations periodically. But in these methods load balancing cannot be performed until the end of a period even if load imbalance has occurred in the middle of the period. In this paper, we present a new threshold load balancing method for workstations which process the jobs with relatively long execution times. The new method decides a proper time to perform load balancing and does perform the balancing right after the detection of the load imbalance. We also show that a static load balancing method with a long period is suitable if the workstations have to deal with the jobs having unpredictable arrival times and relatively short execution times. The performance of the methods presented in this paper is compared with the method without load balancing as well as with the periodic methods. The experiments were done with an iterative data parallel problem called ISING problem. The experimental results show that our methods outperform all the other methods that we compared.