| In this dissertation, we extend caching for two applications. For one application, we extend caching by proposing an architecture for creating a hierarchy of dynamically created caches. For the second application, we extend caching by proposing a cache management policy for a cache that prefetches, predicted data.; In the first part of this dissertation, we address the problem of disseminating fast changing data to a large number of users. Due to the fast changing nature of the data, caching data at clients is not very effective. Additionally, the static nature in which Web caches are configured makes it hard for such caches to adapt to dynamically changing access patterns. In our analysis of the World Cup Soccer data, we found that a large fraction of the clients accessing fast changing data belong to a small fraction of network domains. Our solution to disseminating fast changing data, domain caching, exploits this property by dynamically creating caches when a large number of clients from a domain access the same data. Further, the server instructs all clients in the domain to access data from a selected client, the domain cache .; In the second part of this dissertation, we address the problem of accessing data from a disk farm. Prefetching is one approach to exploit the parallelism provided by a disk farm. Due to the difficulty in predicting future data accesses, prefetching data from disks is hard. Further, all predicted data are guaranteed to be accessed. In this dissertation, we propose and evaluate a cache management policy that integrates caching and prefetching by managing cached data and prefetched data separately. Our cache management policy evaluates the relative values of cached data and prefetched data. Finally, we use this information to identify the least valuable cached block. |
