Replicating Web documents reduces user-perceived delays and wide-area netwo
rk traffic. Numerous caching and replication protocols have been proposed t
o manage such replication while keeping the document copies consistent. We
claim, however, that no single caching or replication policy can efficientl
y manage all documents. Instead, we propose that each document be replicate
d with a policy specifically tailored to it. We have collected traces on ou
r university's Web server and conducted simulations to determine the perfor
mance such tailored policies would produce, as opposed to using the same po
licy for all documents. The results show a significant performance improvem
ent with respect to end-user delays, wide-area network traffic and document
consistency. We also present how these results can be used to build adapti
ve replicated Web documents, capable of automatically selecting the policy
that best suits them. (C) 2001 Elsevier Science B.V. All rights reserved.