Re: [squid-users] Poor performance of squid-2.5 on Redhat 7.2 compared to Solaris 8 on x86

Mahmoud

I am very interested to read that async-io=32 is better than diskd, but I am curious to know how you know this has
increased the performance of requests/hits p/sec this gives over diskd.

Cheers in advance

Dan

On Mon, Jul 01, 2002 at 10:29:15AM +0430, Account For Cache Group wrote:
> I compile most of my squid box with this option, and I have not gotten any
> serious bug yet.
> as you know there are some other technique to reach a better performance
>
> 1. set half_close_client off
> 2. client and server persistent_connection on
> 3. decread fin_timeout timeout (/proc/sys/net/ipv4/tcp_fin_timeout)
> 4. incread read/write tcp buffer ( ..../tcp_wmem, /tcp_rmem)
> 5. disable timestampes
> 6. incread tcp_max_syn_backlog
> 7. and ... you can find them in Securing and Optimizing Linux book
>
> Mahmoud Taghizadeh
>
>
> On Sun, 30 Jun 2002, Erik Rowberg wrote:
>
> >
> > I recompiled last week with --enable-async-io support (after reading
> > a paper by joe cooper) and it is a definite improvement. (RH7.2)
> >
> > Is the aufs no longer concidered buggy, or was async-io the buggy
> > part (according to squid.conf)?
> >
> > Erik
> >
> > > as far as I know, thread mechanism in linux and solaris are different.
> > > instead of "diskd" you can compile squid with --enable-async-io=32 I am
> > > sure that you will get a better performance with this option.
> > > dont forget to change ufs to aufs in cache_dir directive in squid.conf
> > >
> > > Mahmoud Taghizadeh
> > >
> > >
> > > On Wed, 20 Feb 2002, Peter Arnold wrote:
> > >
> > > > Hi,
> > > > I've been working with squid for some time on low end x86/solaris platform.
> > > > I concocted a really slick config on that worked well to impress the powers
> > > > that be such that bought some new IBM PIII/1.13Ghz 512Mb etc etc to replace
> > > > the solaris X86 P200 128Mb etc... woopee! except they wanted Redhat because
> > > > IBM support redhat....
> > > >
> > > > Anyway we have a url we can access that indicates download speed and for
> > > > some reason the solis box ALWAYS leaves the Redhat box for dead....1500Kbps
> > > > vs 200-300Kbps
> > > >
> > > > The same speed applies for the bundled version of Squid in Redhat that is
> > > > 2.4S1.
> > > >
> > > > I recompiled the exact same versions on solaris as redhat with the same
> > > > options....
> > > > ./configure --prefix=/opt/squid-2.5 '--enable-storeio=ufs diskd null' --
> > > > enable-icmp --enable-delay-pools --enable-useragent-log --enable-
> > > > referer-log --enable-snmp --enable-underscores '--enable-auth=basic ntlm' '-
> > > > -enable-basic-auth-helpers=PAM' --enable-ntlm-auth-helpers=NTLMSSP
> > > > except I included LDAP in the helpers on redhat
> > > > Squid is configured on both with a null cache_dir but otherwise is pretty
> > > > stock standard.
> > > > Solaris/Sqiuid still wins by a factor of 5!
> > > >
> > > > Is there any trick to Squid on Redhat that I should know about? Can anyone
> > > > offer any advice on trouble shooting this?
> > > >
> > > > Thanks in advance
> > > > Peter Arnold
> > > >
> > > >
> > >
> >
> >
Received on Mon Jul 01 2002 - 03:22:45 MDT