Designing Algorithms for Dependent Process Failures
Most fault-tolerant algorithms are designed assuming that out of n components, no more than t can be faulty. For example, solutions to the Consensus problem are usually developed assuming no more than t of the n processes are faulty, where “being faulty” is specialized by a failure model. We call this the t of n assumption (also known as threshold model). It is a convenient assumption to make. For example, bounds are easily expressed as a function of t: if processes can fail only by crashing, then the Consensus problem is solvable when t < n if the system is synchronous and when t < 2n if the system is asynchronous extended with a failure detector of the class ◊W. [5.5], [5.1]
Unable to display preview. Download preview PDF.
- 5.2A. Doudou and A. Schiper. Muteness Detectors for Consensus with Byzantine Processes. In Proceedings of the 17th ACM Symposium on Principle of Distributed Computing, page 315, Puerto Vallarta, Mexico, July 1998.Google Scholar
- 5.3M. Hirt and U. Maurer. Complete Characterization of Adversaries Tolerable in Secure Multy-Party Computation. In ACM Symposium on Principles of Distributed Computing, pages 25–34, Santa Barbara, California, 1997.Google Scholar
- 5.4F. Junqueira and K. Marzullo. Consensus for Dependent Process Failures. Technical report, UCSD, La Jolla, CA, September 2002. http://www.cs.ucsd.edu/users/.avio/Docs/Gen.ps.
- 5.5I. Keidar and S. Rajsbaum. On the Cost of Fault-Tolerant Consensus When There Are No Faults-A Tutorial. Technical Report MIT-LCS-TR-821, MIT, May 2001.Google Scholar
- 5.6D. Malkhi and M. Reiter. Byzantine Quorum Systems. In 29th ACM Symposium on Theory of Computing, pages 569–578, May 1997.Google Scholar
- 5.8P. Thambidurai and Y.-K. Park. Interactive Consistency with Multiple Failure Modes. In IEEE 7th Symposium on Reliable Distributed Systems, pages 93–100, Columbus, Ohio, October 1988.Google Scholar