Re: Cacheflow vs. Squid

On Thu, 23 Jul 1998, Fraser Campbell wrote:

> CacheFlow says this technique can boost hit rate to as
> high as 75%, compared with 30% to 40% in most cache systems."

75% is probably for some unusual workload like the one they used for their
performance benchmarking. However, if you have enough bandwidth and 20-40TB
of storage, you can cache "the entire cachable Web" using Squid and wget and
have hit ratios close to 75%. :)

I think that the idea of prefetching soon-to-be-requested pages deserves
consideration (and is certainly not the invention of Cacheflow). However,
from "environmental" point of view, using Cacheflow may be like purchasing a
lot of water and food supplies in Africa, drinking a glass of water, and then
flushing the rest down the toilet.

Note that there is no HTTP-friendly way an origin server can refuse to serve
a Cacheflow box' request (but I am sure a hacked version of Apache will
appear soon). Thus, a Cacheflow box may fetch a lot of data increasing the
load on origin servers and congesting the network, just because its owner has
enough money to pay for the bandwidth. See Carlos reply for the example.

In short, prefetching is a tradeoff between reducing response time and
increasing bandwidth/origin server resources consumption. Currently, HTTP has
no way to limit the appetite of a greedy prefetching algorithm.

Alex.
Received on Fri Jul 24 1998 - 08:35:59 MDT