Title: Virtual I/O Scheduler: an Adaptable Scheduler of Schedulers

Speaker: Seetharami Seelam, The University of Texas at El Paso

Date/Time: Tuesday, October 10, 2006, 10:00 – 11:00 am

Location: CSRI Building/Room 90 (Sandia NM)

Brief Abstract: Disk scheduling algorithms in operating systems often are designed to satisfy a primary data delivery requirement. In addition, these algorithms fail to provide performance isolation among concurrently executing disk-bound applications.  Lack of performance isolation makes it difficult conduct application performance analysis and tuning, hampers forecasting application performance, and may hinder providing quality of service guarantees. In addition, multiple concurrent and conflicting requirements need to be satisfied to support concurrently executing applications while providing fairness in sharing the disk resources.

We have designed a virtual I/O Scheduler (VIOS) that leverages a fair queuing discipline and implements a fair scheduling algorithm that can facilitate satisfying different data delivery requirements of concurrently executing applications by allowing each application to have its own scheduling algorithm. Because this strategy is fair in terms of allocated disk time, it provides predictable disk performance for each application and, therefore, performance isolation among applications.  Being able to predict performance facilitates providing quality of service guarantees.

In this talk we describe the impact of different resource-sharing metrics of conventional fair scheduling on fairness, performance isolation, and application disk performance predictability. We present the fairness properties and analytical and experimental evaluations of our VIOS scheduling algorithm.  We also present the many uses of providing fairness and performance isolation by VIOS for storage virtualization, quality of service objectives, dynamic adaptive I/O scheduler selection, and application performance analysis and tuning.