darrell@sequoia.ucsc.edu (Darrell Long) (07/20/90)
The following technical report is available via anonymous FTP from midgard.ucsc.edu (128.114.134.15). It is in pub/tr/ucsc-crl-90-36.ps.Z -- so be sure to use binary mode transfer. A paper copy of this report is also available. Write to: Jean McKnight Technical Librarian Baskin Center for Computer Engineering & Information Sciences University of California Santa Cruz, CA 95064 Internet: jean@cis.ucsc.edu Resilient Memory-Resident Data Objects Jehan-Francois Paris Department of Computer Science University of Houston Houston, TX 77204-3475 Darrell D.E. Long Computer and Information Sciences University of California Santa Cruz, CA 95064 ABSTRACT Data replication has been widely used to build resilient data objects. These objects nor- mally consist of several replicas stored in stable storage and a replication control protocol manag- ing these replicas. Replicated data objects incur a significant penalty resulting from the increased number of disk accesses. We propose several protocols that improve performance while not significantly impacting the availability of the replicated data object. Our replicated data objects consist of several memory-resident replicas and one append- only log maintained on disk. We analyze, under standard Markovian hypotheses, the availability of these data objects when combined with three of the most popular replication control protocols: available copy (AC), majority consensus voting (MCV) and dynamic-linear voting (DLV). We show that repli- cated objects consisting of n memory-resident replicas and a disk-resident log have almost the same availability as replicated objects having n disk-resident replicas. Keywords: file replication, replicated databases, memory-resident databases, majority consensus vot- ing.