A Cost-Sensitive Adaptation Engine for Server Consolidation of Multitier Applications

Virtualization-based server consolidation is an important technique for cost and energy reductions in data center environments and a key enabler of cloud computing. However, to ensure adequate application isolation and performance, consolidation demands runtime resource reconfiguration, especially of multi-tier services that have dynamic, rapidly changing workloads and responsiveness requirements. While virtualization makes reconfiguration easy through capacity controls, VM replication and migration, their indiscriminate use has performance implications. In this paper, we quantify the costs of Xen-based virtual machine adaptation and show that ignoring those costs can have significant impacts on the ability to satisfy response-time-based SLAs. We address this problem by developing a cost-sensitive adaptation engine that weighs the potential benefits of runtime reconfiguration decisions against their costs. Through extensive experimental results based on live workload traces, we show that the technique is able to maximize SLA fulfillment under typical time-of-day workload variations as well as flash crowds, and exhibits significantly improved transient behavior compared to approaches that do not account for adaptation costs.