CACHING STRATEGIES TO IMPROVE DISK SYSTEM PERFORMANCE

Processor speeds have increased dramatically over the last few years and are expected to continue to double every year, with main memory density doubling every two years. At the same time, our appetite for I/O continues to grow, especially with the emergence of applications such as multimedia and scientific modeling. Although rapid technological advances have doubled disk capacity every 1-1/2 years since 1990, no similar advances are expected to reduce mechanical latency and, thus, the access times of storage devices; access times of main memory and storage devices will likely remain many orders of magnitude apart. As a result, I/O subsystems limit overall system response time and leave the CPU underutilized. This article examines the use of caching as a means to increase system response time and improve the data throughput of I/O subsystems. The authors describe a number of cache parameters that affect cache design and performance, including cache size, various popular line-replacement algorithms, read-ahead and write-to-cache strategies, and cache location. Using simulations, the authors analyze the effectiveness of three cache replacement algorithms for workload traces from four different systems. Using an algorithm with low implementation overhead, known as segmented least recently used (SLRU), the authors find that in certain circumstances SLRU halves the cache size required by other popular caching strategies.