1. Cost-Aware WWW Proxy Caching Algorithms Pei Cao University of Wisconsin-Madison Sandy Irani University of California-Irvine Proceedings of the USENIX Symposium on Internet Technologies and Systems Monterey,California,Dec ember 1997 University of Yuan Ze System Lab Mike Tien miketien@syslab.cse.yzu.edu.tw

2. Outline 1.Existing Result 2.Existing Document Replacement Algo 3.Comparison between Existed Proxy Replacement Algo 4.GreedyDual-Size Algo 5.Performance Comparison 6.Summary 7.Conclusion

3. 1.Existing Result Three differences between Web caching and memory page 1.web cache is variable-size caching. 2.web objects take different amounts of time to download,even if they are of the same size. 3.access streams seen by the proxy cache are from ten to thousands of user. Optimal Result:For a sequence of requests to uniform size blocks of memory,the simple rule of evicting the block whose next request is farthest in the future.(web hit ratio 50% [WASA96]).

4. 2.Existing Document Replacement Algorithms LRU ad:simplicity disad:not include file size and latency LFU Size LRU-Threshold — LRU,except documents larger than certain threshold size. Pitkow/Recker — removes the least-recently-used document,except if all documents are accessed today,in which case the largest one is removed. Lowest-Latency-First — tries to minimize average latency by removing the document with the lowest download latency first. Hybrid--F= Lowest Relative Value – takes into account locality,cost and size of a document.

5. 3.Comparison between Existed Proxy Replacement Algorithms Size >> LFU,LRU-threshold,and Pitkow/Recker [WASAF96].Size >> LRU in most situation. LRU >> Size in terms of byte hit rate. In most cases,LFU << LRU. In terms of minimizing latency,Hybrid >> LLF. LRV >> both LRU and Size in terms of hit ratio and byte hit ratio. One disadvantage of both Hybrid and LRV is their heavy parameterization. But,the studies offer no conclusion on which algorithm a proxy should use.

6. 4.GreedyDual-Size Algorithm

7. 5.Performance Comparison Hit Rate &Byte Hit Rate

10. Reduced Latency

11. Network Cost

13. 6.Summary High Hit Ratio -> GD-Size(1) High Byte Hit Ratio -> GD-Size(packet) If network latency don ’ t change over time or change slowly ->GD-Size(hops) or GD-Size(weightedhops) Small cache size -> GD-Size(1)

14. 7.Conclusion GD-Size combines locality,cost and size Only optimize one performance measure at a time.Looking into how to adjust the algo. To optimize more than one performance measures Plan to study the integration of hint-based prefetching with the cache replacement algo.