Events - Colloquia & Seminars

CCIS Colloquium Spring 2008

Beyond crash faults: Tolerating byzantine faults in database systems

Speaker: Ben Vandiver

Date: Friday, March 21, 2008

Talk: 10:30 a.m., 366 WVH

Abstract

According to commercial and open source database vendor archives, over 50% of verified bugs result in non-crash faults (e.g., incorrect answers, data corruption, etc). Yet the database fault tolerance community has focused almost exclusively on tolerating crash faults. I will present the first practical system for tolerating arbitrary (Byzantine) faults in database systems. The system, called HRDB, uses middleware to provide a single-copy serializable view of a replicated database, and relies on heterogeneous replicas to provide increased fault tolerance. Central to my performance claims is a novel concurrency mechanism that allows highly concurrent execution of a transaction processing workload on heterogeneous replicas.