Dynamic Contents Clustering for the Disk I/O Enhancement in Web Proxy Servers

摘要

With the increasing popularity of World Wide Web, network traffic due to HTTP has dramatically increased. World Wide Web proxies are used to reduce user-perceived latency and wide-area network traffic by caching frequently accessed web objects. However, recent researches show that the bottleneck of a web proxy cache server is the disk I/O. In general, proxies are running on a UNIX file system that has the different disk I/O workload characteristics than those of proxies. In this paper, we study the problems of caching web objects on the disk and suggest a Dynamic Contents Clustering for the disk I/O enhancement, which preserves the locality of the HTTP requests by translating it into a cluster in disk space. Clusters are generated dynamically based on web access patterns and they are a basic unit of the disk I/O. It reduces seek and rotation delay of the disk system by increasing the unit of the disk I/O. Additionally, we use BUDDY disk layout system in a file for reduction in file meta-data management overhead and the fiat name space in a file system. We evaluate the performance of our approach by a mix of trace-driven simulation and experimental evaluation.