Review – On the scale and performance of cooperative Web proxy caching

Internet proxy caching is a common approach for improving client browsing
performance and reducing the bandwidth requirement of Internet Service
Providers. While there are clear benefits, it is not trivial to determine
where the limits of this approach are, and how cooperation between caches
can provide the best cost performance. Wolman et al. use observations that
the hit rate of a proxy is a function of the size of the population it
manages, along with several trace experiments, to determine that the largest
benefit for cooperative caching is achieved for relatively small
populations.

This paper explains that there is little motivation to expanding the study
of highly scaleable cooperative web caching, because the benefits of
increasing a population flatten out sharply after only a few thousand
clients. To arrive at this conclusion, many important questions needed to be
answered: what is the best performance possible, what is the impact of
population size, what are the effects on latency and bandwidth as population
is increased, etc. All of these topics are considered with trace analysis.

The authors note that the population level at which cooperative caching
works efficiently is limited by document cacheability and that research
should be done in this area. However, it seems that since the writing of
this paper, objects on the Internet have become less cacheable. With the
advent of AJAX and generally increasingly dynamic Internet content, it would
be interesting to study the current level of cacheable content. For example,
Google Maps and Google Mail both use AJAX (Yahoo and Microsoft are in the
process of developing AJAX mail interfaces). This is not cacheable content
and its presence on the internet is growing.

Perhaps the most interesting statement in this paper is that, 'unpopular
documents are universally unpopular.' While this seems like a trivial
observation, it has fundamental implications for cooperative caching. That
is, assuming a large and diverse enough population, if the object is not
available at the local cache it is likely not available at a nearby cache.
Received on Thu Oct 20 2005 - 10:18:40 EDT