Re: [squid-users] squid + Intel STL2 and slow performance after few hours

In a dual PIII 866 with 1GB RAM, Linux 2.2.19 without problems!

[]-------------------------------------------------------------[]
  Pedro Ribeiro
  Online: http://www.isel.ipl.pt/~pribeiro/
  IRC(PTnet) Nick: PAntMaR
  e-Mail: Spamm: trap@net.ipl.pt
               Personal: pribeiro@net.ipl.pt
               IPLNet/ISELNet Admin: helpdesk@net.ipl.pt
  PGP Key Available in keyservers.
  PGPFP: 6430 6CA7 7642 36D6 718B A884 FDF0 E8F1 30EA 1D19
[]-------------------------------------------------------------[]
----- Original Message -----
From: "Mitesh P Choksi" <mitesh.choksi@stcl.com>
To: <squid-users@squid-cache.org>
Sent: Wednesday, July 25, 2001 3:32 PM
Subject: RE: [squid-users] squid + Intel STL2 and slow performance after few
hours

> An update here,
>
> Kernel 2.4.2 and kernel 2.4.6 both don't work. I have tried to find out
any
> problems with Interl STL2 motherboard with 2 1GHz processors and 1GB RAM
> but have not got any answers. Intel sent a doc to check the MPS version
set
> to 1.1 and not 1.4 which could give the problem. The BIOS does not have
any
> facility to change that version.
>
> So as it stands, the server is next to useless, unless after some time I
use
> it as my personal web cache server. :(
>
> Something to do with kernel as I have realised that when i stop squid and
> have large 'memory cache' even untarring takes up all that cpu utilisation
> even ftp download is slow.
>
> I will keep trying but any input is highly appreciated. (anyone using SLT2
> server board with linux+squid?)
>
> Mitesh
>
> -----Original Message-----
> From: Mitesh P Choksi [mailto:mitesh.choksi@stcl.com]
> Sent: 25 July 2001 12:37
> To: 'Chemolli Francesco (USI)'; squid-users@squid-cache.org
> Subject: RE: [squid-users] PLEASE HELP: Large cache deployment and squid
>
>
> In case this gives some idea. As soon as i started the L4 switching to
point
> to squid server, the system usage, i.e. squid usage jumped to and hit
99%.
> This was about 12 requests/second.
>
> procs memory swap io system
> cpu
> r b w swpd free buff cache si so bi bo in cs us
sy
> id
> 0 0 0 0 118180 51944 685940 0 0 0 48 119 135 0
5
> 95
> 0 0 0 0 118176 51944 685940 0 0 0 0 101 145 0
0
> 100
> 1 0 0 0 118176 51944 685940 0 0 0 2 111 134 0
3
> 97
> ----------- requests startes going in
> 2 0 0 0 118044 51944 685960 0 0 4 0 247 38 0
82
> 18
> 1 0 0 0 117848 51944 686044 0 0 5 18 221 36 0
77
> 23
> 2 0 0 0 117644 51944 686132 0 0 16 45 241 38 0
80
> 20
> 2 0 1 0 117628 51944 686140 0 0 0 6 228 3 0
99
> 1
> 2 0 1 0 117284 51944 686244 0 0 1 22 219 43 0
80
> 20
> ----------- requests stopped to keep the server alive
> 2 0 0 0 117256 51944 686264 0 0 0 58 193 34 0
84
> 16
> 0 0 0 0 117244 51944 686264 0 0 0 6 119 126 0
11
> 89
> 1 0 0 0 117228 51944 686272 0 0 0 26 115 139 0
5
> 95
> 1 0 0 0 117216 51944 686280 0 0 0 18 116 142 0
4
> 96
> 0 0 0 0 117188 51944 686292 0 0 0 26 117 135 0
5
> 95
> 1 0 0 0 117172 51944 686292 0 0 0 13 124 140 0
5
> 94
> 1 0 0 0 117168 51944 686296 0 0 0 10 125 131 0
7
> 93
> 0 0 0 0 117168 51944 686296 0 0 0 1 111 135 0
2
> 98
>
> May be the server board chipset is not supported, and that high IO is
giving
> a problem. When the memory cache is low, then no problems, it works like a
> breeze.
>
> Mitesh
>
> -----Original Message-----
> From: Mitesh P Choksi [mailto:mitesh.choksi@stcl.com]
> Sent: 25 July 2001 12:31
> To: 'Chemolli Francesco (USI)'; squid-users@squid-cache.org
> Subject: RE: [squid-users] PLEASE HELP: Large cache deployment and squid
>
>
> Hi Chemolli,
>
> Thanks for the direction. I tried with --enable-time-hack and
> with --disable-time-hack. but did not find any difference. The only thing
I
> see here is that as soon as the pagecache increases to 400+MB even a
single
> request can drive the squid CPU utilization to 5% or more.
>
> Can you please send me the command line that you have used on teh system.
I
> have also read something about SMB and non-RAM limiting pagecache giving
> some problem. Maybe that it's the kernel that will require tuning rather
> than squid.
>
> My question is how do I find out what is the problem. Is there anything in
> squid that can let me know this.
>
> Moreover, does gettimeofday() and poll() fire even if I have /dev/null in
> the cache logs. If so than i can spare some CPU time by disabling logs.
>
> Thanks for any input in advance.
>
> Mitesh
>
> -----Original Message-----
> From: Chemolli Francesco (USI) [mailto:ChemolliF@GruppoCredit.it]
> Sent: 25 July 2001 10:03
> To: 'mitesh.choksi@stcl.com'; squid-users@squid-cache.org
> Subject: RE: [squid-users] PLEASE HELP: Large cache deployment and squid
>
>
> > PLEASE HELP
> >
> > I have been trying to post a questoin about very large cache
> > to the list but
> > have not succeeded.
> >
> > Anyways, I have the following situation, and don't know if
> > anybody has faced
> > it
> > as well.
> >
> > 2 * 1GHz PIII
> > Intel Server Motherboard (ServerWorks PCI compiled in the kernel.)
> > 1GB RAM
> > cache_mem = 2MB
>
> How come this little cache_mem? You have plenty of RAM, use it...
>
> > pagecache = 2 15 75
> > SCSI HDD
> > cache_dir = 3 * 13GB on different disks
> > linux kernel = 2.4.2-2 smp (recompiled to remove all
> > unnecessary stuff and )
> > squid version = 2.4STABLE1 (from redhat rawhide) (had to
> > install openssl as
> > well)
> >
> > After 12 hours of use even 150+ request / minute (yes
> > minutes) squid takes
> > up 70%+ CPU.
>
> I have half your CPU speed and manage 150+ req/sec (280 during peak-time).
> I don't get those kinds of loads.
>
> > It has also happened to me with 2.3STABLE4 (from default
> > rh7.1 linux cd)
> > kreiserfsd taking up 10% cpu at times
>
> Check out Joe Cooper's performance tuneups (they are referenced
> from somewhere on the Squid site).
> The only case I had of such a CPU usage was when using a broken
> --enable-time-hack.
> I'd suggest that you rebuild squid from scratch on your system (from the
> tarball)
>
> > memory cache goes to 555MB+ that's what vmstat shows. squid
> > takes up ~20MB
> > as shown by top
>
> the 500Mb part is not strange, since Linux allocates all unused RAM to
that.
>
> > Strace showed me the following after the 12 hour check (poll %time is
> > surprising)
> > 100.11 -2088636.745941 -440 2907882 poll
>
> Squid is spinning like crazy in the select loop.
> Might be an effect of --enable-time-hack.
>
> > All documents I read says Squid does not require more
> > processing power, yet
> > I gave it 2 x 1GHz. Is that a problem?
>
> That's much more than you'll need unless you do more than
> 400 req/s.
>
> --
> /kinkie
>
>
Received on Wed Jul 25 2001 - 10:06:37 MDT