NGS: Providing and Maximizing Quality of Service in Utility Computing Servers

North Carolina State University, Raleigh NC

Investigators

Abstract

This work proposes new run-time, operating system, and performance modeling techniques that enable Quality of Service (QoS) through providing performance guarantee and contention minimization for user jobs in future utility computing servers. Utility Computing is an IT business model where users outsource computing to a vendor that manages computing resources and charges its customers based on the actual computing that they use. Utility computing servers will likely simultaneously run many user jobs with varying performance guarantee requirements. Intellectual Merit: The need to provide performance guarantee in high performance servers. The proposed work seeks to address. The proposed work will make revolutionary advances to the current state of the art technology in batch job submission system, OS job scheduling, and performance modeling. More specifically, the proposed approach consists of three new components: (1) Job Admission Strategy that accepts or rejects jobs based on the servers' ability to meet the requested QoS, (2) Job Scheduling Strategy that boosts the system throughput by minimizing the contention among multiple jobs, and (3) Contention Prediction Model that predicts the degree of contention and its impact on job execution times for a given job schedule. Broader Impact: The proposed work takes QoS issues from their traditional area in computer networks to a new area in servers, run-time systems and operating systems.