Quality of service on multicore systems using cache control techniques

Abstract

Data center environments employ powerful multicore servers on which multiple applications with varied latency requirements are coscheduled to maximize resource utilization Recent trends in virtualisation and cloud computing technologies have resulted in multiple virtual machines being coscheduled on multicore platforms Providing Quality of Service (QoS) guarantees for performance of applications running in such environments has been an important area of research in the past decade QoS is usually provided by managing one or more shared resources used by applications. This work explores managing one such shared resource the shared cache to provide QoS guarantees. This work classifies a program as a high priority program or a low priority program where QoS is provided to the high priority program at the expense of performance degradation of the low priority program The work proposes three methods to provide QoS (a) a static approach based method that provides QoS on cache usage (b) a dynamic approach based method that provides QoS on cache performance and (c) a dynamic approach based method that provides QoS on overall performance The static approach based method provides QoS on cache usage In this method cache usage requirement of a high priority program is specified as the fraction of time during which the entire cache is available for its exclusive use Core level cache control technique was used to manage the shared cache The evaluations showed that the method can be used in providing QoS and helped identify the runtime parameters at which the method was most effective newlineHowever the method did not consider individual phases of execution within a program and resulted in unnecessary performance loss of the low priority program X The dynamic approach based method resolves the issues of the static approach based method and provides QoS on cache performance