Detecting Resource Contention in Virtualized Systems

Abstract

Public and private cloud computing environments typically employ virtualization methods to consolidate application workloads on a single server for gaining cost and energy benefits. Contention among applications for shared server and virtualization resources can have a significant impact on application performance. Such contention can lead to resource bottlenecks that can especially be problematic for interactive applications such as Web servers that need to support fast response times for user requests. Previous work suggests monitoring hardware platform specific performance metrics for detecting such contention. This research shows that such metrics are not always sufficient for detecting contention, especially for highly concurrent interactive applications such as Web servers. A novel software probe based approach is presented for addressing this limitation. We show that the probe imposes a low overhead and is remarkably effective at detecting common performance degradations that can occur in environments featuring both interactive and batch style workloads.