On-Demand Prefetching Heuristic Policies: A Performance Evaluation

International audiencePrefetching is a basic mechanism in the World Wide Web that speculates on the future behaviour of users to avoid the response delays. The relatively new requirement of the instantaneous response in some interactive services like On-Demand applications fuelled the need for ways to represent and reason about the challenging problem of prefetching control and performance evaluation. We study this challenging problem under a network protocol that adopts the simultaneous prefetching with equal-shared bandwidth, and in prefetching situations in which the controller seeks to reach a Zero-Cost system state as quickly as possible. Within this context, our first contribution is providing the backbone of a new paradigm for the performance evaluation of the On-demand prefetching policy. This backbone consists of our previously developed prefetching control model; the PREF-CT model and our previously developed optimal control algorithms; the ONE-PASS and the TREE-DEC algorithms. Our second contribution is developing the prefetching heuristic algorithm: the RBP. Compared to the optimal prefetching policies, the prefetching policies computed by our heuristic algorithm the RBP show significant performance in terms of the user’s latency and the bandwidth utilization