Web Caching and Replication

12.8. Prediction Algorithms

Every prefetching system faces the problem of predicting a client's future accesses based on information about the past. Prediction algorithms differ in the level of past usage detail they consider and how far in the future they try to predict. Simpler, coarser-grain information may miss some prediction opportunities but is easier to collect and process. Some algorithms consider only coarse-grain history information such as aggregate object popularity. These algorithms do not use access patterns in their predictions. Using the terminology of Section 6.5, they do not exploit spatial locality of reference. Many other algorithms attempt to reach conclusions of the form, "if a client has accessed a sequence of objects ...