Performance analysis of an object-oriented approach to parallel query evaluation

Abstract

An object-oriented representation of databases which takes advantage of the structure of logical representation and an object-oriented algorithm that evaluates conjunctive queries in a massively parallel environment are reported and analyzed. The object-oriented representation is used to implement data level parallelism as data is encapsulated as an object. By distributing these objects in a database, parallel evaluation is achieved by a global pattern matching approach which is different from the conventional tuple-by-tuple, operation-by-operation evaluation strategy. This algorithm overcomes the data dependency imposed by shared variables which is a major problem in achieving AND-parallelism. Simulation results for 2 way and 3 way joins are analysed according to the effect of database size and join size. Results tend to show that the scheme proposed performs better for queries with large number of AND conjuncts.<>