Enhanced file prefetching for delivery on demand
Paul F. Reynolds, Timothy Highley · 2005
Delivery on demand could make any program immediately available to a user without tedious installations or trips to a software store. Vendors could use it to ensure that the latest, most secure versions of software are in use. Despite its apparent benefits, delivery on demand has not been available in the general user area, due in part to the size of large software programs and connections too slow to deliver the programs quickly. File prefetching is a key technology for enabling delivery on demand, and could be even more useful when coupled with multiple channels of communication between the source and the user. Cost-benefit analysis is an approach to file prefetching that has been demonstrated to be effective under several system models. In this dissertation, I discuss the challenge of file prefetching and how cost-benefit analysis can be extended to be more effective. I have investigated predictive file prefetching, identifying an algorithm for probabilistically optimal prefetching as well as an algorithm that simulations show to be an improvement over any other approach in the literature. I also used cost-benefit analysis as a basis to develop the first algorithm for predictive file prefetching across multiple channels with different monetary costs.