Hi,
This might looks more like a squid-user problem but I think
explanations from the developers' perspective should give me more
insights. Here is the problem:
I use httperf to request a single nocachable file. Squid can only
sustain less than 200 req/s. (The file size is 4k and I'm using a slow
Linux box. Squid is configured with default settings). But Apache runs
on the same machine can sustain about 600 req/s if the requests are
sent directly to Apache. I can imagine that Squid needs about twice as
many sockets as Apache (because of client-side and server-side) but
that should not degrades Squid's performance so much. Also Squid use
event driven model which technically should overperformance Apache
which uses process pool model. So, what's the bottleneck, from
developers' point of view?
BTW, for some historical reason I'm using Squid-2.3, if that makes big
difference.
I actually found somebody asked this question way back to 2000 but
there was only one reply to that.
Thanks a lot for your help.
-- ChengduReceived on Thu May 01 2003 - 11:23:39 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:19:51 MST