Scheduling BoT applications in grids using a slave oriented adaptive algorithm

Abstract

Efficient scheduling of Bag-of-Tasks (BoT) applications in a computational grid environment reveals several challenges due to its high heterogeneity, dynamic behavior, and space shared utilization. Currently, most of the scheduling algorithms proposed in the literature use a master-oriented algorithm, in which the master is the only responsible for choosing the best task size to send to each slave. We present in this paper a different approach whose main originality is to be slave-oriented, i.e. each slave locally determines, from a set of initial runs, which workload size is more adapted to its capacities and notifies the master of it. Finally, we show some measurements comparing our algorithm with other three well-known scheduling algorithms using the SimGrid toolkit. © Springer-Verlag Berlin Heidelberg 2004.