Resource Conservation Using Virtualization
Research Lead: Amin Vahdat, Associate Professor, Computer Science and Engineering, UC San Diego
Much CS research is at the forefront of technologies and architectures for deploying and managing emerging data center computing architectures as well as work in managing the increasingly critical energy resources of the computing infrastructure . A logical abstraction layered on top of a set of physical machines, virtual clusters, exploits virtual machine technology to more efficiently share computational resources between competing application demands while ensuring fault isolation.
Usher is a complete implementation of a virtual cluster management middleware in daily use at UCSD. Usher exposes dynamic levels of parallelism, and will support the distinctive needs of the GreenLight Instrument's computing environments and it applications, including large-scale scientific computing as well as network services with its resource allocation mechanisms and higher level policies for cluster resources. We will study dynamically migrating virtual machines to appropriate architecturally configured physical machines to achieve network locality on campus given the GreenLight Instrument's proposed switching and routing. Today's power provisioning, on a per-rack basis, must be done by worst-case analysis, rather than based on operational information, so, for this Instrument, we will also study power consumption and make it part of Usher. In many situations, it may be possible to achieve a factor of 2 or more increase in density by provisioning for the average case draw and dynamically migrating to virtual machines for power affinity.
Technical White PapersVirtualization 2010