As discussed in the preceding chapter when the cache is full we require to evict one or more cached objects to make place for the new incoming cached object. This is done by the replacement policy. By having a good replacement policy it is possible to get better performance from the cache system. A good replacement policy is required to select a cached object for removal which will least affect the performance of the cache. The object selected for removal could be one has not be accessed for a long time and hence probably will not be accessed again.
It turns out that the best performance is achieved by the policies that take into account the size of the documents evicted. LRU-MIN is a policy derived from LRU (least recently used) that tries to minimize the number of documents evicted by applying LRU only to the documents whose size is above some threshold. The threshold is adaptive: if not enough space has been released, the threshold is lowered and the policy is applied again. This policy outperforms every other removal policy. One that gives a lower but comparable performance is LFU (least referenced file removed first). There are three key differences between Web caching and Conventional paging problems.
Most policies are some form of LRU. The Advantage of LRU is it's simplicity. It's drawback is it does not consider file sizes & latency. Hence LRU may not give best results when applied to web caching.