mod-os@sdcsvax.uucp (01/12/87)
-- The following is the abstract for a technical report (CS-092) that has just been released. If you are interested in obtaining a copy, please address inquiries to: Technical Report Librarian Department of Computer Science and Engineering, C-014 University of California, San Diego La Jolla, CA 92093 DL -- Block-Level Consistency of Replicated Files John L. Carroll Computer Science Group Department of Mathematical Sciences San Diego State University San Diego, CA 92182 Darrell D. E. Long Jehan-Francois Paris Computer Systems Research Group Department of Computer Science and Engineering University of California, San Diego La Jolla, Ca 92093 ABSTRACT To increase the availability and reliability of files the data are often replicated at several sites. The usual method is to treat files as logi- cal entities and to replicate on a per file basis. This can lead to unnecessary complications for the implementor in trying to preserve file system semantics. We investigate the construction of a reliable device. Such a device would appear to the file system as an ordinary block-structured device, but would be implemented as a set of server processes on several sites. This allows us to provide repli- cation while leaving the operating system kernel and the file system unchanged. The regular structure of the block-level replication environment allows the use of con- sistency control algorithms that are simpler and less network intensive. We present three algo- rithms for maintaining file consistency in a block-level replication environment. The first is a majority consensus voting algorithm that recov- ers blocks only when they are required for data access; the second is a variant of the available copy scheme modified for replication at the block level; the third is a naive version of the avail- able copy scheme that does not maintain any failure information. Each scheme is evaluated in terms of availa- bility and network traffic. While block-level replication is shown to allow improvements in the network traffic burden incurred by voting, the available copy schemes are shown to have better availability and require significantly less traffic than the voting schemes. The naive avail- able copy variant proposed here is shown to be the algorithm of choice. --