Multiple Level Grouping File System for Web Proxy Server

• Main Authors: I-Fan Chen, 陳逸帆
• Other Authors: Tsozen Yeh
• Format: Others
• Language: en_US
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/61100583562848898109

碩士 === 輔仁大學 === 資訊工程學系 === 93 === Many organizations run proxy server on personal computer, and most of these proxy servers store data in file systems managed by operating system. Many of these file systems are designed for the workload of workstation. However the workload of proxy server is very different from that of workstations, and therefore these proxy server cannot take good usage of the properties of proxy documents locality through these file system. Multiple level grouping file system (MGFS), on the other hand, is a file system embedded in proxy server, and so that MGFS can take advantage of the document locality of proxy server. MGFS takeovers the whole document cache machinism including memory and disk, and the replacement scheme. MGFS divides memory cache buffer into four sections: hot section, medium section, cold section, and grouping section. Hot section contains small documents that most probably be cached hit, medium section stores small documents probably will be accessed again, and cold section caches small documents that not been accessed for a while and larger documents that may cost a lot of memory space. Grouping section groups small documents by three levels. At the first level of grouping section documents are clustered by the URL based hashed key, at second level documents are assorted by the key generated from the clients IP and time, and at the last level documents are classified by the size. As the improvement of personal computer and the growth of Internet bandwidth, the contains of web sites are getting more lush, and the size of documents embedded in web pages is getting larger. Since computers are getting more powerful and much economic. Proxy servers are capable to pursue lower response time and higher byte cached hit ratio. In MGFS we present a mechanism that try to pursue better performance and higher cached hit ratio. In order to achieve the goal of lower response time and higher cached hit ratio. MGFS follows two policies. First, keep as many documents that most probably be cached hit in memory as possible. Second, minimize the impacts of large documents as much as possible. And therefore, after analyzing the workload of proxy servers. Large documents in memory are stored in cold section in order not to pollute hot and medium section, and stripe read/write scheme are employed for large documents. Small cached hit documents are migrated to hot section and if the document was found in disk, other documents grouped to the same cluster in disk will be prefetched into medium section. Besides of prefetched documents, the first fetched documents are put to medium section too, for there are too many documents that only accessed for once. If the hot section is full, the least recent used document in hot section will be migrated to medium section, and so that if the medium section is full the least recent used document in medium will be swept to cold section. The same, if the cold section is full the least recent used document in cold section will be swept out, and if this document is a small one it will be put into grouping section, or just stripe write into disk if it is a large one. In our experiments, MGFS performs well not only in hit ratio but also the response time. We compare MGFS with UCFS which is also a file system specified to proxy servers, and found that the response time is shorter than UCFS even though MGFS caches documents under 1 MB and UCFS only caches documents under 256 KB.