Re: Performance problem on Solaris 2.5.1

Ery,
        I forgot to add - for our 640MB system, I have the cache_mem value
set at 196.

Rich.

> Ery,
> We used to have a very simlar setup here, but the problems seen
> before have gone away by the upgrade to Solaris 2.6 and by using select()
> instead of poll(). I did not use the GNU malloc library, and the disks were
> arranged as a Raid 1 partition across 3 disks controllers in an SSA. Our
> bottlenecks occurred in disk I/O and these changes solved that.
> At the moment we're happily getting around 50-100,000 connections
> per hour on the 2.6 machine.
> You could try decreasing the meomry value here so the squid process
> doesn't use too much and stop swapping? Others may have better ideas.
>
> Rich.
>
> > Dear,
> >
> > We have Squid V 1.1.21 running on Ultra Sparc 2, RAM 512MB, Harddisk
> > 24G, with OS Solaris 2.5.1. We encounter performance degradation after
> > running the squid for about 10-12 Hours, we have already follow the
> > instruction that we got from the mailing list digest, but we still
> > encounter the problem.
> > Right now we use GNUMalloc package for the malloc routine and GCC
> > 2.7.2.1 as the compiler. The statistic form the cache information show
> > the average connection for 1 hour about 50.000-80.000 connections. There
> > are 20.000 to 30.000 users we have that connected to the internet via
> > that solaris box (actually we have 3 identical machine).
> > The main configuration of squid.conf show like this :
> >
> > cache_swap 17000
> > cache_mem 200
> >
> > httpd_accel 202.134.0.227 80
> > httpd_accel_with_proxy on
> > httpd_accel_uses_host_header on
> >
> > we running the machine on transparent proxy mode inconjunction with
> > Cisco 7500 router. IP filter 3.2.7 have been installed on the solaris
> > box as the redirector. The main problem we got for 10 Hours period, the
> > memory shown by the vmstat always decrement (it's seem memory leak) into
> > smaller number (we notice almost 70% of the swap memory used). Paging
> > activity is so high, and the disk activity shown by the iostat at the
> > level 100%!.
> > Could be the memory allocation problem solve by using any other malloc
> > library, for example DL malloc ? although we got problem on the
> > compiling the DL malloc library for our squid 1.1.21. Or, the Metadisk
> > daemon we use can eat all the io process ? (the iostat show very high
> > number), avg disk service about 20ms ...
> > Is there anybody can help us, we really desperately to get the squid
> > machine fine tuned :(
> >
> > Thanks
> >
> > Ery Punta Hw
> > Product Manage
> > TELKOMNet
> > e-mail:phunta@telkom.net.id
> > phone:62-21-3160500
> >
>
>
> --
> ..Blue O "Smoke me a kipper,
> Skies.. //\/ I'll be back for breakfast."
> \/\ ..Must
> ...../ Dash..
>

-- 
..Blue         O            "Smoke me a kipper,
  Skies..    //\/            I'll be back for breakfast."
            \/\  ..Must
         ...../    Dash..