Parallel transferable uniform multi-round algorithm for achieving minimum application turnaround times for divisible workload

Abstract

A parallel transferable uniform multi-round (PTUMR) scheduling algorithm is proposed for mitigating the adverse effect of the data transmission time by dividing workloads and allowing their parallel transmissions to distributed clients from the master in a network. The performance of parallel computing using the master/worker model for distributed grid computing tends to degrade when handling large data sets due to the impact of data transmission time. Multiple-round scheduling algorithms have therefore been proposed to mitigate the effects of the data transmission time by dividing the data into chunks that are sent in multiple rounds so as to overlap the time required for computation and communication. However, standard multiple-round algorithms assume a homogeneous network environment with uniform link transmission capacity, and as such cannot minimize the turnaround time effectively in real heterogeneous network environments. The proposed PTUMR algorithm optimizes the size of chunks, the number of rounds, and the number of workers to which data is to be transmitted in parallel, and is shown through performance evaluations to mitigate the adverse effects of data transmission time between the master and workers significantly, achieving turnaround times close to the theoretical lower limits. © Springer-Verlag Berlin Heidelberg 2005.