Database creation and/or reorganization over multiple database backends.

Abstract

To create a record in a database, one uses the INSERT command. However, in the Multi-Backend Database System (MBDS), the insert command only inserts one record at a time. When creating very large databases consisting of thousands or millions of records, the use of the INSERT command is a time-consuming process. Once a database is created, some of the records of the database may be tagged for deletion. MBDS uses the DELETE command to tag these records. Over some period of time, those records tagged for deletion should be physically removed from the database. Hence, removing tagged records is in essence creating new databases from untagged records. In this thesis, we present a methodology to efficiently create very large databases in gigabytes on parallel computers and to reorganize them when they have been tagged for deletion. Specifically, we design a utility program to by-pass the system's INSERT command, to load the data of the database directly onto disks and create all the necessary base data and meta data of the database.