Files in this item
|(no description provided)|
|Title:||Database support for electronic CAD application: Performance and architecture|
|Author(s):||Chu, Sheauyin Iris|
|Doctoral Committee Chair(s):||Winslett, Marianne|
|Department / Program:||Computer Science|
|Degree Granting Institution:||University of Illinois at Urbana-Champaign|
|Abstract:||This research addresses the problems in obtaining efficient database support with good functionality for electronic CAD applications. We believe that performance is a deciding factor in the acceptance of DBMSes as a replacement of file systems or home-grown databases in ECAD tools and that many DBMS design and implementation choices can have a major impact on the performance of the resulting system.
Fast loads, fast stores, and fast in-memory references are all important performance considerations. The first part of this thesis assessed empirically the impact of different architectural features on the performance of database systems by running benchmarking experiments on CAD tools using various DBMSes as the underlying storage system. In these experiments, the DBMS was used to retrieve and store data only, representing a loosely-integration between the DBMS and the CAD tool. Results of these experiments indicate that architectural choices such as access methods, clustering, transfer granularity, process architecture, application overhead all have great impact on database performance for load and store operations. In addition, the tightly-coupled OO1 benchmark was ported to several commercial DBMSes to investigate tightly-coupled interactions between the DBMS and the CAD tools. The kind of client-server architecture was found to have the most significant impact.
A detailed study of different client-server architecture choices shows that a 'page-to-page' architecture will permit an implementation where most routine reference following (i.e., where the referenced data is in memory and appropriately locked) is handled by virtual memory hardware to eliminate expensive software overhead. Page-to-page systems, however, have a number of limitations such as unacceptable low concurrency for high-contention data pages and difficulties in supporting fine-grained authorization. The second part of this research focuses on how to improve the functionality of page-to-page server. We showed how to use hardware assisted locking of minipages (subdivisions of a page), which is already available on many well-known platforms, to overcome these limitations. Simulation results show that the minipage approach performs essentially as well as the standard page-to-page server when contention for data is light, while offering the ability to do fine-grained authorization. In addition, the minipage approach performs much better than the standard page-to-page server when there is much contention for data between clients.
|Rights Information:||Copyright 1995 Chu, Sheauyin Iris|
|Date Available in IDEALS:||2011-05-07|
|Identifier in Online Catalog:||AAI9522095|