Title
"Parallel resource sharing: Shortcomings, Cheats, and Perspectives"


Abstract

Parallel computing resources, such as commodity clusters, need to be shared among users so as to justify hardware and power costs. This sharing is typically done today via batch schedulers. Batch schedulers unfortunately have several shortcomings: they lead to non-deterministic queue waiting times, do not attempt to optimize formally defined objective functions, and only enable suboptimal resource utilization. In this presentation we will discuss approaches for coping with these shortcomings, some of which are stigmatized as "cheating", some of which attempt to define radically new resource-sharing models and technologies.

Prof. Henri Casanova

Associate Professor at Information and Computer Sciences Department - University of Hawaii at Manoa