Continual Computation Policies for Utility-Directed Prefetching

Eric Horvitz

Decision Theory & Adaptive Systems Group
Microsoft Research
Redmond, Washington 98052-6399

Click here to access the paper in postscript format.
Click here to access the paper in pdf format. (Alternate: access pdf at ACM site.)

Abstract:

People accessing documents via the Internet typically experience latencies in retrieving content. We discuss continual-computation policies that dictate strategies for prefetching into cache portions of documents that a user may wish to review later. These utility-directed prefetching strategies maximize the expected utility of idle network resources that are available frequently, yet sporadically during the review of documents accessed from the World Wide Web. We present policies based on alternate utility models for assigning value to having immediate access to content and discuss means for coupling the methods with probabilistic models that predict a user’s interests and access behavior.

Keywords: Continual computation, prefetching, caching, network bandwidth, cost--benefit analysis, decision theory, decision-theoretic methods, flexible computation.

In: Proceedings of the Seventh ACM Conference on Information and Knowledge Management, Bethesda MD, November 3-7 1998, pp. 175-184. ACM Press: New York.

Author Email: horvitz@microsoft.com