Effective Resource and Workload Management in Data Centers.

摘要

The increasing demand for storage, computation, and business continuity has driven the growth of data centers. Managing data centers efficiently is a difficult task because of the wide variety of datacenter applications, their ever-changing intensities, and the fact that application performance targets may differ widely. Server virtualization has been a game-changing technology for IT, providing the possibility to support multiple virtual machines (VMs) simultaneously. This dissertation focuses on how virtualization technologies can be utilized to develop new tools for maintaining high resource utilization, for achieving high application performance, and for reducing the cost of data center management.;For multi-tiered applications, bursty workload traffic can significantly deteriorate performance. This dissertation proposes an admission control algorithm AWAIT, for handling overloading conditions in multi-tier web services. AWAIT places on hold requests of accepted sessions and refuses to admit new sessions when the system is in a sudden workload surge. To meet the service-level objective, AWAIT serves the requests in the blocking queue with high priority. The size of the queue is dynamically determined according to the workload burstiness.;Many admission control policies are triggered by instantaneous measurements of system resource usage, e.g., CPU utilization. This dissertation first demonstrates that directly measuring virtual machine resource utilizations with standard tools cannot always lead to accurate estimates. A directed factor graph (DFG) model is defined to model the dependencies among multiple types of resources across physical and virtual layers.;Virtualized data centers always enable sharing of resources among hosted applications for achieving high resource utilization. However, it is difficult to satisfy application SLOs on a shared infrastructure, as application workloads patterns change over time. AppRM, an automated management system not only allocates right amount of resources to applications for their performance target but also adjusts to dynamic workloads using an adaptive model.;Server consolidation is one of the key applications of server virtualization. This dissertation proposes a VM consolidation mechanism, first by extending the fair load balancing scheme for multi-dimensional vector scheduling, and then by using a queueing network model to capture the service contentions for a particular virtual machine placement.