Data Dissemination on the Web: Speculative and Unobtrusive Institute for Advanced Computer Studies

The rapid growth of the Web results in heavier loads on server/network and in increased latency experienced while retrievingWeb documents. Internet traffic is further aggravated by its burstiness, which complicates the design and allocation of network components. Bursty traffic alternates peak periods with lulls. 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 it is explicitly aware of current HTTP traffic loads so as to be unobtrusive. This paper highlights several design trade-offs and details the problem of server arbitration among several candidate documents. We present a theoretical analysis of arbitration and validate it by extensive simulation on server logs, in which we calculate latency experienced by clients. Perfect traffic shaping during peak periods is observed and substantial latency improvements for non-dynamic documents are reported over pure on-demand strategies.