(An) implementation of hardware filter processor for the relational data base machine

Abstract

A hard-ware data filter is implemented. It filters necessary tuples to perform join operation. To check the tuple has probabilities of participation in join operation, it use two hash bit arraies. And to enhance performance of the filter, it adopts a dualism. So, in this filter two disk are used and each of which has half of tuple of all relations. This filter has two data filter path also, and each of which does crossreference each other to check the joinability. To join two relations they are flown simultaneously through this two path. This filter is composed with 4 parts, disk I/O module, tuple comparator unit, buffer unit and hash unit. The tuple comparator unit does simple selection operations. The buffer unit temporarily stores tuples which is judged not to participate in join operation until disk reading is ended. The hash unit use two sets of hash functions and hash bit arraies to decrease collision ratio. Filter structure and its operation is presented.