The performance advantage of applying compression to the memory system

Abstract

The memory system stores information comprising primarily instructions and data and secondarily address information, such as cache tag fields. It interacts with the processor by supporting related traffic (again comprising addresses, instructions, and data). Continuing exponential growth in processor performance, combined with technology, architecture, and application trends, place enormous demands on the memory system to permit this information storage and exchange at a high-enough performance (i.e., to provide low latency and high bandwidth access to large amounts of information). This paper comprehensively analyzes the redundancy in the information (addresses, instructions, and data) stored and exchanged between the processor and the memory system and evaluates the potential of compression in improving performance of the memory system. Analysis of traces obtained with Sun Microsystems' Shade simulator simulating SPARC executables of nine integer and six floating-point programs in the SPEC CPU2000 benchmark suite yield impressive results. Well-designed compression schemes may provide benefits in performance that far outweigh the extra time and logic for compression and decompression. This will be more so in the future since the speed and size of logic (which will be used to perform compression/decompression) are improving and are projected to improve at a much higher rate compared to those of interconnect (which will be used to communicate the information), both on-chip and off-chip.