USENIX 2005 Annual Technical Conference, General Track Abstract
Pp. 7588 of the Proceedings
CHAMELEON: A Self-Evolving, Fully-Adaptive Resource Arbitrator for Storage Systems
Sandeep Uttamchandani, IBM Almaden Research Center; Li Yin, University of California, Berkeley; Guillermo A. Alvarez and John Palmer, IBM Almaden Research Center; Gul Agha, University of Illinois at Urbana-Champaign
Abstract
Enterprise applications typically depend on guaranteed
performance from the storage subsystem, lest they fail. However,
unregulated competition is unlikely to result in a fair, predictable
apportioning of resources. Given that widespread access protocols and
scheduling policies are largely best-effort, the problem of providing
performance guarantees on a shared system is a very difficult one.
Clients typically lack accurate information on the storage system's
capabilities and on the access patterns of the workloads using it,
thereby compounding the problem. CHAMELEON is an adaptive arbitrator for
shared storage resources; it relies on a combination of self-refining
models and constrained optimization to provide performance guarantees
to clients. This process depends on minimal information from clients,
and is fully adaptive; decisions are based on device and workload
models automatically inferred, and continuously refined, at run-time.
Corrective actions taken by CHAMELEON are only as radical as warranted by
the current degree of knowledge about the system's behavior. In our
experiments on a real storage system CHAMELEON identified, analyzed, and
corrected performance violations in 3-14 minutes--which compares very
favorably with the time a human administrator would have needed. Our
learning-based paradigm is a most promising way of deploying
large-scale storage systems that service variable workloads on an
ever-changing mix of device types.
- View the full text of this paper in HTML and PDF.
Until April 2006, you will need your USENIX membership identification in order to access the full papers. The Proceedings are published as a collective work, © 2005 by the USENIX Association. All Rights Reserved. Rights to individual papers remain with the author or the author's employer. Permission is granted for the noncommercial reproduction of the complete work for educational or research purposes. USENIX acknowledges all trademarks within this paper.
- If you need the latest Adobe Acrobat Reader, you can download it from Adobe's site.
|