Information Services and Process Innovation Lab

The data dependability project focuses on evaluating storage system dependability, designing storage solutions to meet users' business requirements, and determining how to recover systems after disasters. Customers provide dependability business goals expressed as financial penalty rates: $/hr of outage time (unavailability) and $/hr of recent data loss, specified in terms of how many recent updates are lost during recovery. These penalty rates can then be used to evaluate the financial ramifications of different storage solutions and recovery schedules, and to choose the most appropriate one.

Selected publications:

• Designing dependable storage solutions for shared application environments (PDF, 103KB), Shravan Gaonkar, Kimberly Keeton, Arif Merchant, and William Sanders, Proc. of Intl. Conf. on Dependable Systems and Networks (DSN), June 2006, Philadelphia, PA.

  This paper shows how to design dependable storage systems employing multiple data protection techniques for shared application environments.

• On the road to recovery: restoring data after disasters (PDF, 311KB), Kimberly Keeton, Dirk Beyer, Ernesto Brau, Arif Merchant, Cipriano Santos and Alex Zhang, Proc. of ACM European Systems Conference (EuroSys), April 2006, Leuven, Belgium.

  This paper describes the problem of data recovery scheduling, introduces recovery graphs to represent alternative approaches for recovering data systems, presents optimization and heuristic approaches for solving the problem, and evaluates these methods using realistic storage system designs and workloads.

• A framework for evaluating storage system dependability (PDF, 154KB), Kimberly Keeton and Arif Merchant, Proc. of DSN, June 2004, Firenze, Italy.

  This paper presents a framework for modeling the dependability of commonly used data protection techniques, and describes how to compose these techniques to determine the dependability of the overall storage system.

• Designing for disasters (PDF, 305KB), Kimberly Keeton, Cipriano Santos, Dirk Beyer, Jeffrey Chase, and John Wilkes, Proc. of 3rd USENIX Conf. on File and Storage Technologies (FAST), pp. 59-72, March 2004, San Francisco, CA.

  This paper frames the problem of disaster planning, where the goal is to minimize the overall costs of a dependable storage system, including up-front outlays and post-disaster penalties. It presents a mixed integer program optimization specification of the problem, and solves this specification using an off-the-shelf solver to choose the best solution for a variety of business requirements.

» Download the latest Adobe Acrobat Reader