A benchmarking methodology for the centralized-database computer with expandable and parallel database processors and stores

Description

In this paper a benchmarking methodology for a new kind of database computers is introduced. The emergence in the research community and in the commercial world of this kind of database computer (known as the multiple-backed database computers), were each computer system is configured with two or more identical processors and their associated stores for concurrent execution of transactions and for parallel processing of a centralized database spread over separate stores, is evident. The motivation and characterization of the multiple-backend database computer are first given. The need and lack of a methodology for benchmarking the new computer with a variable number of backends for the same database or with a fixed number of backends for different capacities are also evident. The measures (benchmarks) of the new computer are articulated and established and the design of the methodology for conducting the measurements is then given. Because of the novelty of the database computer architecture, the benchmarking methodology is rather elaborate and somewhat complicated. To aid our understanding of the methodology, a concrete sample is given herein. This sample also illustrates the use of the methodology. Meanwhile, a CAD system which computerizes the benchmarking methodology for systematically assisting the design of test databases and test-transaction mixes, for automatically tallying the design data and workloads, and for completely generating the test databases and test-transaction mixes is being implemented.

Subjects