I want to get some successful story for squid on Sun Sparc platform.
We used to service several thousand users through parallel "parent cache" forward proxies deployed on E280. These had the advantage of our choice to deploy local "child cache" servers at the various remote intranet sites; parent proxies would accept HTTP/ICP queries only from a handful of internal "child cache" clients, so the ugly TCP and HTTP problems (e.g. broken browsers sending malformed queries) would be mostly filtered out before requests made it to the parents. Another advantage of a hierarchical approach is that we could deploy sets of parents in different cities with different ISPs, with the children being ICP-aware of multiple parents in multiple sites. If any one site were to be nuked (logically or literally), children automatically fail-over to the remaining site(s).
We are running Squid 2.5STABLE13 on Solaris 10 X86 on two V20z for several 1000 Users.
. . .
> I would guess that Linux currently is the best performaning OS > if you apply the epoll patch. Can you give us some more details why you think that Linux will perform better?
As much as I dislike Linux, there is something to be said for the scalability improvements provided by "the epoll patch". Solaris 10 has enhanced event handling(/dev/poll and the new event ports), however I am not aware of any initiative to enable Squid to take advantage of these Solaris features. Kevin
