Re: Not serving pages well, Solaris7?

hillel@learn.co.za wrote:

> I've been told by users that my squid http accelerator, that I've been
> working on for
> the last month, running squid-2.3.STABLE2 (should I upgrade to STANBLE3 ?)
> is caching the pages but not serving them well.
>
> I've done ./configure --enable-heap-replacement --enable-dlmalloc
> --enable-snmp --enable-time-hack

why the time hack?

> I've also set /etc/system as I'm using Solaris 7 on Sun 420
>
> *priority paging
> set priority_paging=1

That is usually a good thing to do.

>
> *filedescriptors
> set rlim_fd_max=4096
> set rlim_fd_cur=256

Did you have a look into your system on how well it is performing
from several points of view? Performance of HTTP and DNS as reported
by the cachemgr, of HTTP/TCP level as reported by sending the host
probes, TCP kernel level as reported by "netstat -s", on memory
level as reported by top and vmstat, on io level as reported by
iostat -xe, etc. The Solaris performance guru is Adrian Cockroft,
and his are the books (and SWOL pages) to be read.

There is only very little TCP/IP tweaking left to do with Sol7, but
did you increase the socket buffer sizes? Is slow_start_initial
set to 2 (especially important with BSDish clients like Windoze)?
Increasing the TCP connection hash size might help, if the server
is really busy.

If all the above don't give some clear indications, you'll might
need to look into TCP traces - which is a pain on busy servers.

-- 
Le deagh dhùrachd,
Dipl.-Ing. Jens-S. Vöckler (voeckler@rvs.uni-hannover.de)
Institute for Computer Networks and Distributed Systems
University of Hanover, Germany; +49 511 762 4726