Abstract:
The need to transfer data through a PCI-E (Peripheral Component Interconnect Express) bus is one of the key characteristics of GPU and multicore coprocessors programming, which is considered as a bottleneck for a number of applications. This paper focuses on evaluating the efficiency of data compression for optimizing the data transfer between main memory and Intel Xeon Phi for database applications. Three compression methods are evaluated: LZSS (Lempel-Ziv-Storer-Szymanski), Null Suppression, and RLE (Run-Length Encoding). An implementation of these methods for Intel Xeon Phi coprocessors is described. It is shown experimentally that these compression methods can be used to increase the efficiency of database processing under certain conditions imposed on the data under treatment. It is also shown that, when a compression method allows one to process data without decompression, such a processing procedure can additionally increase the efficiency of this method.