[CSC2231] Paper Review: On the Scale and Performance of Cooperative Web Proxy Caching

This paper investigates the performance of cooperative web proxy caches
in two ways. First, it studies traces of web activities to determine the
benefits of cooperative caches for organizations of different sizes.
Then, it crafts an analytic model to determine the performance of large
scale cooperative caches. Overall, the paper shows that cooperative
caches work well for small populations.

My intuition aligns with the findings of the paper. If two organizations
have different access patterns, cooperative caching is not useful at
all. If an organization is large, it has the resources to build large
web proxy and this proxy is sufficient without needing to cooperate. So,
cooperative caching only works with the organizations that have similar
access patterns and have limited storage at their web proxies. In fact,
as storage space price drops significantly these years, the storage web
proxies for even small organizations can grow and be self-sufficient.
This raises the question whether cooperative caching is still any useful.

The cooperative caching technique reminds me of NUMA machines. In NUMA
machines, each processor can access its own cache memory or memory from
other processors. Accessing data at remote memory of course requires
more cycles. I believe that there are significant research efforts
investigating ways to migrate data across different processor's cache
memory to achieve overall performance improvements. Such techniques may
be applicable to balance cooperative web caches as well to improve the
overall hit rates.
Received on Thu Oct 20 2005 - 10:08:15 EDT