Cost-aware Dynamic Provisioning for Performance and Power Management

Dynamic provisioning of server boxes to applications entails aninherent performance-power trade-off for the service provider, a trade-off thathas not been studied in detail. The optimal number of replicas to bedynamically provisioned to an application is ultimately theconfiguration that results in the highest revenue. The serviceprovider should thus dynamically provision resources for an applicationonly as long as the resulting reward from hosting more clients exceeds its operational costs for power and cooling.We introduce a novel cost-aware dynamic provisioningapproach for the database tier of a dynamic content site. Our approachemploys Support Vector Machine regression for learning a dynamicallyadaptive system model. We leverage this lightweight on-line learningapproach for two cost-aware dynamic provisioning techniques. The firstis a temperature-aware scheme which avoids temperature hot-spotswithin the set of provisioned machines, and hence reduces cooling costs.The second is a more general cost-aware provisioning technique usinga utility function expressing monetary costs for both performance and power.