Title: Discovering Spatial Locality in WWW Access Patterns using Data Mining of Document Clusters in Server Logs

Authors: Azer Bestavros

Date: September 10, 1997

Abstract:

In this paper, we introduce the notion of a ``document cluster'' in
WWW space as a generalization of the notion of a ``cache line'' in
linear memory address space.  Through the analysis of Web server logs,
we show evidence of the spatial locality of reference in WWW access
patterns and present an implementation of an efficient data mining
algorithm that discovers document clusters.  We show preliminary
simulation results that quantify the benefits of using document
clusters for file allocation on server disks, as well as for purposes
of prefetching into server cache/main memory.