tradeoff between cache memory and bandwidth

Figure 10 extends the results of the previous two figures, dividing the marginal savings in bytes transferred from the central server by the cache size. Minimizing this ratio for most zones given these two days of queries implies an optimal timeout is between 64 and 512 seconds.

Other considerations also will affect the preferred timeout. In particular the timeout may likely be a per-document parameter rather than a per-cache parameter, since some documents have a very low mean time between modification (MTBM) and thus caching them for more than a few seconds, or perhaps even at all, would be imprudent. These are parameters which will require considerable site-specific empirical investigation from studies such as this one. Rather than the results of this particular set of logs, we emphasize in this study the methodology and importance of web log characterization to tune caching for system efficiency.