Peter J. Varman Professor of ECE and CS

Virtualization technologies and server consolidation have the potential to reduce both fixed and operating costs within a data center, and to reduce total energy consumption. Good QoS control is key to a supportable economic cost model for such services, and for establishing clients’ trust in a remote platform. An increasingly pressing problem in this regard is the lack of effective QoS support for IO resources, which render even existing CPU and memory controls ineffective.

In this talk, we will summarize our recent results in resource provisioning and scheduling for QoS support in storage IO servers. We discuss the use of workload shaping in a server environment to reduce provisioned capacity by filtering workload bursts and providing graduated QoS guarantees. We will also discuss dynamic scheduling to support generalized QoS requirements in both centralized array and decentralized cluster environments, and their VM-based evaluation.

Wednesday, October 14, 2009
11:00 a.m. - McMurtry Auditorium, Duncan Hall

2009 ECE Affiliates Conference
Building Stronger Connections