|
|
|
@inproceedings{DBLP:conf/vldb/BanerjeeH78,
author = {Jayanta Banerjee and
David K. Hsiao},
editor = {S. Bing Yao},
title = {Performance Study of a Database Machine in Supporting Relational
Databases},
booktitle = {Fourth International Conference on Very Large Data Bases, September
13-15, 1978, West Berlin, Germany},
publisher = {IEEE Computer Society},
year = {1978},
pages = {319-329},
ee = {db/conf/vldb/BanerjeeH78.html},
crossref = {DBLP:conf/vldb/78},
bibsource = {DBLP, http://dblp.uni-trier.de}
}
Database machines are special-purpose devices that are expected to perform the common data management operations efficiently. In this paper, we attempt to show how a relational database can be supported on a specific database machine, known as the database computer (DBC), with good performance.
The DBC employs modified moving-head disks for database storage. To achieve high-volumed accessing, the read-out mechanisms of the moving-head disks are made into tracks-in-parallel. To provide content-addressable search, the disk controller is incorporated with a set of microprocessors, corresponding to the tracks of a cylinder. In this way, not only can an entire cylinder of data be accessed in one disk revolution, but relevant data which satisfies,the user request can also be found and output in the same revolution.
To minimize the number of cylinders involved in a database access, some structural information about the database is maintained in a blockoriented content-addressable memory made of charge-coupled devices (CCDs). Furthermore, clustering and security mechanisms are a part of the hardware features provided by the DBC.
With cylinder-oriented content-addressable database store, block-oriented content-addressable structure memory and several functionally specialized components, the DBC can achieve one or two orders of magnitude of performance improvement over the conventional computer in database management. Also, a possible twofold increase in database storage requirement as compared to a conventional implementation is adequately offset by one or more orders of magnitude reduction in storage for structural information.
The purpose of this paper is to analyze these performance issues. By using the DBC for supporting relational databases, the size of the relational software is considerably reduced. Specifically, the query optimizer of conventional systems is now rendered unnecessary. In comparison with a conventional implementation of a relational system, the DBC has been found to contribute larger performance gains. These gains are tabulated in the paper. All these tend to demonstrate that the DBC in particular and database machines in general can indeed contribute to an appreciable improvement in database management.
Copyright © 1978 by The Institute of Electrical and Electronic Engineers, Inc. (IEEE). Abstract used with permission.
