Evaluating the Presence of a Victim Cache on an Arm Processor

Mobile processor is a CPU designed to save power. It is found in mobile computers and cell phones. A CPU chip, designed for portable computers, is typically housed in a smaller chip package, but more importantly, in order to run cooler, it uses lower voltages than its desktop counterpart and has more sleep mode capability. A mobile processor can be throttled down to different power levels and/or sections of the chip can be turned off entirely when not in use. ARM is a 32-bit reduced instruction set computer (RISC) instruction set architecture (ISA). The relative simplicity of ARM processors makes them suitable for low power applications. Hence ARM processors account for approximately 90% of all mobile 32-bit RISC processors. Today, mobile processors are expected to run complex, algorithm-heavy, memory-intensive applications which were originally designed and coded for general-purpose processors. Due to this we see a huge impact of the memory latencies on the execution time of applications. To reduce this impact and serve this kind of applications, the relative complexity of ARM processors has increased in the last decade by the inclusion of traditional methods like multiple issue of instructions, out-of-order instruction execution and large, associative caches. Victim Caching is another method which can be used to reduce the execution time and is currently not incorporated in the ARM processors. This method was proposed by Norman P. Jouppi in his paper “Improving Direct-Mapped Cache Performance by the Addition of a Small Fully-Associative Cache and Prefetch Buffers”. Victim Cache is defined as an extension to a direct mapped cache that adds a small, secondary, fully associative cache to store cache blocks that have been ejected from the main cache due to a capacity or conflict miss. These ejected blocks are likely to be needed again so storing them in the secondary cache should increase performance and reduce the execution times. Therefore for the Master\u27s project we re-implemented the SimpleScalar simulator for an ARM processor by incorporating the impact of Victim Cache. This re-implementation of the ARM simulator gave a significant improvement in the performance when various applications of MIBench benchmark suite were run on this simulator. It is observed to have a reduction of 1.93% in the number of clock cycles used and increase in the hit rate of Level 1cache by 2.7% over various Level 1 cache and Victim cache configurations on an average. It is also observed that the benefit of Victim cache increases as the size of Level 1 cache decreases and the performance boost obtained by the processor in presence of a Victim cache is comparable to the performance obtained when a Large, Associative Level 1 cache is used. Hence, incorporation of Victim Cache to an ARM processor is highly advantageous to the current generation of Mobile processors instead of using a Large, Associative Level 1 cache