Providing QoS through service level agreements in high performance computing

Abstract

This papers shows how the state of the art in high-performance computing job scheduling can be augmented to provide not only best-effort based services or, at best, trivial quality of service parameters, but differentiated service levels. A solution is proposed using long-term service level agreements (SLAs), in contrast to per-job SLAs that are normally used. An architecture for the the implementation of SLA-based job scheduling is presented and a workload trace, enhanced with service level data, is simulated in order to analyze the impact of service levels on key functions. © Civil-Comp Press, 2011.