Challenges and Hopes for Scheduling Applications on Desktop Grids
Prof. Henri Casanova
University of Hawaii at Manoa
The efficient scheduling of applications onto distributed compute resources is known to be theoretically difficult and has been studied for several decades. Desktop grids pose the additional challenge that the common case is for resources to be reclaimed or to fail without notice. To go beyond traditional throughput maximization for large and long-running applications on desktop grids, scheduling algorithms must be developed that tolerate and even exploit the stochastic nature of desktop grid resources. In this presentation we will review the general approach in this area for developing and validating such algorithms. We argue that this approach, while popular and appealing, reduces the usefulness and the impact of obtained results on the field. It is thus necessary to "take a step back" and study the desktop grid scheduling problem in a theoretical framework that is perhaps less realistic but that is more tractable. The hope is that these results can provide fundamental understanding upon which sound scheduling algorithms and practices can be developed.