(A) study on the architecture of on-the-fly filter for the relational database machine

Abstract

In general, scanning records is a necessary prerequisite to database operations. So CPU time for applying scanning predicate is saved if records are checked during reading time from disk. In this thesis, a hardware filter called on-the-fly filter, which is useful for sequential scanning of records in database storage systems is developed. Its architecture is an extension of Content Addressable File Store (CAFS) to be appropriate to Wisconsin Storage System (WISS). The access method for our hardware filter is based on WISS, and it corresponds to the sequential scan of original WISS. The incorporation of the filter and its access method is verified by showing that the data stream is not lost but is correctly accessed through the filter by the access method. For the verification we have designed a simulator for our on-the-fly filter and thus the access method runs on the simulator. We expect that this hardware filter gives an improvement of selection operations for relational database machine with lower cost and convenient access by the upper level of access method.