Extended Abstract (7 pages)
Official ACM version
Local copies: Postscript (830KB) PDF (68KB)
The rapid growth of traffic on the World-Wide Web results in heavier loads on networks and servers and in increased latency experienced while retrieving web documents. This paper presents a framework that exploits idle periods to satisfy future HTTP requests speculatively and opportunistically. Our proposal differs from previous schemes in that speculative dissemination always gives precedence to on-demand traffic, uses ranged requests for improved performance, and can be implemented over a connectionless transport. The protocol uses bounded and little server state even as the workload was increased and it is resistant to erroneous estimates of available bandwidth. Substantial latency improvements are reported over pure on-demand strategies.
In Performance Evaluation Review, Volume 28, Number 2, September 2000, pages 43-49. Presented at the Performance and Architecture of Web Servers (PAWS) Workshop, held in conjunction with ACM SIGMETRICS 2000: International Conference on Measurement and Modeling of Computer Systems, Santa Clara, CA, June 17-18.
A longer version of this paper is available as a technical report.
Back to Brian Davison's publications