UC Santa Cruz

We introduce the aggregating cache, and demonstrate how it can be used to reduce the number of file retrieval requests made by a caching client, improving storage system performance by reducing the impact of latency. The aggregating cache utilizes predetermined groupings of files to perform group retrievals. These groups are maintained by the server, and built dynamically using observed inter-file relationships. Through a simple analytical model we demonstrate how this mechanism has the potential to reduce average latencies by 75% to 82%. Through trace-based simulation we demonstrate that a simple aggregating cache can reduce the number of demand fetches by almost 50%, while simultaneously improving cache hit ratios by up to 5%.