Web Caching and Replication

9.4. Caching on a Global Scale: Proxy Pruning

None of the approaches considered so far is suitable for the scale of the global Internet. Centralized directories have limited geographical scalability because of synchronous directory probes. Replicated directories have limited scalability due to high storage overhead, which grows proportionally with the number of proxies times the individual cache size. In broadcast hierarchies, scalability is limited because either the fan-out or the depth of the hierarchy must increase with scale. Increasing the fan-out causes an exponential increase in the likelihood of query timeouts; increasing the depth increases the latency on a miss and reduces the hit rate. URL hashing does not take network proximity into ...