Project Description
An important method for customers to reduce the cost of data storage is to use fewer devices more efficiently, placing diverse workloads onto a small number of storage systems. This creates the potential for unexpected performance interferences, such as when a load spike on one application causing severe performance degradation on a mission-critical application. The quality of service (QoS) project uses active workload control to avoid and relieve system congestion so that the most important applications maintain the performance they need.
Work in the early stages of the project (see, for example, the paper at SRDS2003) showed the feasibility of active storage workload management to achieve QoS goals. Current research builds upon the core technology to make it directly usable in storage products.
The key challenge is to offer a straightforward and usable formulation to the customer, while coping with the complexity of storage systems under the covers. For example, techniques like write-back caching and lazy copy operations mean that different kinds of I/O streams can have widely different response times, disk cost per I/O, and relationship between the two. The algorithm that automatically adjusts of controls must offer sensible and acceptable behavior in any combination of workloads.
Selected Publications
- David D. Chambliss, Guillermo A. Alvarez, Prashant Pandey, Divyesh Jadav, Jian Xu, Ram Menon, Tzongyu P. Lee, "Performance virtualization for large-scale storage systems", in Proceedings of the Symposium on Reliable Distributed Systems (SRDS), 2003, Florence, Italy.
Selected Patents
- U.S. Patent No. 7,228,354 B2, METHOD FOR IMPROVING PERFORMANCE IN A COMPUTER STORAGE SYSTEM BY REGULATING RESOURCE REQUESTS FROM CLIENTS, D. Chambliss and D. Jadav
People
- David Chambliss
- Rui Zhang
- Ohad Rodeh
