Replicated data management in the Gamma database machine

To help ensure availability of the system in the event of processor and/or disk failures, the Gamma database uses an availability technique termed chained declustering. Like Tandem´s mirrored disk (MD) mechanism and Teradata´s interleaved declustering mechanism, chained declustering uses both a primary and backup copy of each relation. All three systems can sustain the failure of a single processor or disk without suffering any loss in data availability. It is shown that chained declustering provides significantly better performance than mirrored disks in the event of a disk failure. In addition, when no failure has occurred, chained declustering can potentially provide better performance than MDs for input/output (I/O) bound applications