We introduce two versions of the GreedyDual-Size algorithm, GD-Size(1) and GD-Size(packets). GD-Size(1) sets the cost for each document to 1, and GD-Size(packets) sets the cost for each document to 2+size/536 (that is, the estimated number of network packets sent and received if a miss to the document happens). In other words, GD-Size(1) tries to minimize miss ratio, and GD-Size(packets) tries to minimize the network traffic resulting from the misses.
Figure 6: Hit ratio and byte hit ratio comparison of the algorithms.
Figure 6(a) shows the average hit ratio of the four groups of traces under LRU, Size, LRV, GD-Size(1), and GD-Size(packets). The graphs from left to right show the results for Boston University traces, Virginia Tech traces, DEC-U1 traces and DEC-U2 traces, respectively. Figure 6(b) is a simplified version of 6(a) showing only the curves of LRU and GD-Size(1), highlighting the differences between the two. Graph (c) shows the average byte hit ratio for the four trace groups under the different algorithms.
The results show that clearly, GD-Size(1) achieves the best hit ratio among all algorithms across traces and cache sizes. It approaches the maximal achievable hit ratio very fast, being able to achieve over 95% of the maximal hit ratio when the cache size is only 5% of the total data set size. It performs particularly well for small caches, suggesting that it would be a good replacement algorithm for main memory caching of web pages.
However, Figure 6(c) reveals that GD-Size(1) achieves its high hit ratio at the price of lower byte hit ratio. This is because GD-Size(1) considers the saving for each cache hit as 1, regardless of the size of document. GD-Size(packets), on the other hand, achieves the overall highest byte hit ratio and the second highest hit ratio (only moderately lower than GD-Size(1)). GD-Size(packets) seeks to minimize (estimated) network traffic, in which both hit ratio and byte hit ratio play a role.
For the Virginia Tech traces, LRV outperforms GD-Size(packets) in terms of hit ratio and byte hit ratio. This is due to the fact that those traces have significant skews in the probability of references to different sized files, and LRV knows the distribution before-hand and includes it in the calculation. However, for all other traces where the skew is less significant, LRV performs worse than GD-Size(packets) in terms of both hit ratio and byte hit ratio, despite its heavy parameterization and foreknowledge.
LRU performs better than SIZE in terms of hit ratio when the cache size is small (less or equal than 5% of the total date set size), but performs slightly worse when the cache size is large. The relative comparison of LRU and Size differs from the results in [WASAF96], but agrees with those in [LRV97].
In summary, for proxy designers that seek to maximize hit ratio, GD-Size(1) is the appropriate algorithm. If both high hit ratio and high byte hit ratio are desired, GD-Size(packets) is the appropriate algorithm.
