| In this dissertation we address the problem of evaluating performance of web caching replacement algorithms. By developing a model that is able to generate adjustable, representative and realistic workloads of proxies, we provide a fair evaluation environment and investigate how certain algorithms perform with different user access patterns.; The second aspect of our approach is that we present a new caching algorithm called "Bucket" to optimize the use of disk throughput which is always the bottleneck in today's user-server communication. The "Bucket" algorithm considers both the reference frequency and the fetching cost when making a replacement decision. We apply the Lagrange multiplier theory to obtain the optimal solutions for hit rate, byte hit rate, latency and disk throughput. We then utilize the trace-driven simulation to prove the performance gain.; Furthermore, we develop an analytic cache model to estimate the system bus throughput that is efficient to analyze variant design options without real implementation. |
