Abstract

In the engineering and scientific marketplaces, the workstation-server model of computing is emerging as the standard of the 1990s. Implementing an object-oriented database system in this environment immediately presents the design choice of how to partition database functionality between the server and workstation processors. To better understand the alternatives to this fundamental design decision, we analyze three different workstation-server architectures, evaluating them both qualitatively and through benchmarking of prototypes. The three approaches are labeled object server, in which individual objects pass between the server and workstation, page server, in which a disk page is the unit of transport and the server buffers pages, and file server, where whole pages are transferred as well, but they are accessed directly by the workstation process via a remote file service (namely, NFS). We built prototypes of all three architectures, using a stripped-down version of the WiSS storage system as a starting point. To compare the performance of the prototypes, and to experiment with sensitivity to data placement and cache sizes, we developed our own object manager benchmark, the Altair Complex-Object Benchmark (ACOB). This benchmark supports experiments that vary both clustering (inter-object locality) and smearing (intra-object locality). The test suite of benchmarks includes queries for scanning the entire database and traversing and updating complex objects.

Citations