Processing-In-Memory of Data Filter On Compressed Data

Abstract

The data filter is essential in data-centric applications, but it requires moving large data sets to the processing units. One approach to tackle such a hassle is data compression by using lightweight compression methods such as dictionary encoding. In this paper, we exploit the idea of Processing-In-Memory (PIM) data filters directly over compressed data. The initial experiments show noticeable speed-ups of over 2x against the AVX512 architecture.