I haven't seen this problem. I'm able to run tests at up to about 200
reqs/sec from pretty modest poly machines (450-650 MHz, with 256MB RAM).
I think they would go quite a bit higher than that (probably 400-500
if CPU usage is any indicator), actually, but I haven't had need to go
faster except for very short periods.
We're using RH 6.2 and 7.0 on our test boxen, though, so maybe something
has changed since then. Before that I used FreeBSD on our test
machines, since much of the polygraph suite didn't work well on Linux
back then.
Our boxes are setup to compile Squid with high numbers of file
descriptors, but I think Polygraph detects FD's at runtime (and besides,
unless something else is slowing it down, polygraph doesn't use nearly
as many FDs as Squid under load). But if you're hitting problems at 40
reqs/sec, something else is badly wrong. I would probably suspect
network problems...
Rajesh Agrawal wrote:
> Hi,
>
> I am getting good performance(140Req/sec) if i run tests with polygraph
> installed on freebsd.
>
> However, the same cache and same workload file if tested
> with polygraph running on Linux (Redhat 6.0) gives only
> 40 req/sec. I am not getting too many errors in report
> file.
>
> #errno count count% explanation
> 263 40 100.00 "premature end of msg body"
>
> Is there any known way to optimize Linux to get good
> performance ?
>
> Thanks,
> Rajesh Agrawal
--
Joe Cooper <joe@swelltech.com>
Affordable Web Caching Proxy Appliances
http://www.swelltech.com
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:18 MDT