Cloud Workload Characterization and Profiling for Resource Allocation

Abstract

Cloud providers aim to efficiently deliver diverse services on demand to users. Recently, they coined the idea of an auction-based market for their resources with the goal of increasing the total revenues. To address the challenge of scheduling and pricing, we build usage profiles for cloud workloads and predict future demands. In this paper, we first present a new methodology to categorize workloads according to their resource usage. We employ a modified hierarchical clustering algorithm that gives us three demand profiles for batch jobs designated as low, medium and high. After that, we extract the number of arrival requests per time for each group. The methodology presented here provides insights to cloud service providers in optimizing resource allocation and improving profits.