Network-aware HEFT scheduling for grid

Abstract

We present a network-aware HEFT. The original HEFT does not take care of parallel network flows while designing its schedule for a computational environment where computing nodes are physically at distant locations. In the proposed mechanism, such data transfers are stretched to their realistic completion time. A HEFT schedule with stretched data transfers exhibits the realistic makespan of the schedule. It is shown how misleading a schedule can be if the impact of parallel data transfers that share a bottleneck is ignored. A network-aware HEFT can be used to yield a benefit for Grid applications. © 2014 Muhammad Murtaza Yousaf and Michael Welzl.