Optimizing resource topologies of workload in the cloud by minimizing consumption and maximizing utilization while still meeting service level agreements

Abstract

Reducing the total cost of ownership (TCO) and meeting the service level agreements (SLA) of a computing service or application in the Cloud remains a challenge. The traditional method of meeting SLA is to size production systems based on expected peak workloads, thus this leads to very low compute resource utilization. Our experience shows typical average utilization rates of 35 - 40\%, topped only by IBM Mainframe systems with utilization rates of up to 80 - 85\%. In the `Cloud' with an inherent multi-tenant environment resource utilization rates of below 40\% translates into a prohibitive cost factor and therefore into a noncompetitive business model. In this paper we discuss the dynamics of a policy based automated resource orchestration approach for servicing multi-tenant workloads in the Cloud. Our aim is to adapt system topologies dynamically based on changes in workload at runtime. The condition for being able to achieve these goals are: 1) a composable, self-tuned infrastructure (smart IaaS), 2) automatic coordination and provisioning of managed middleware (smart PaaS) and last but not least 3) elastic, self-driven orchestration of software defined workloads and system topologies (smart SaaS).