Re: [squid-users] Squid performance issues

From: Felipe W Damasio <felipewd_at_gmail.com>
Date: Wed, 27 Jan 2010 19:55:11 -0200

  Hi Mr. Lauro,

  Attached are the files you asked for:

  iostat -dx 1 11
  vmstat 1 11
  netstat -s

  Both with a baseline (ie, at a non-troubled time :-)), and at a
moment of pressure.

  On the baseline, "/usr/bin/time squidclient http://www.amazon.com"
took 0.03s, and on the pressure files, the same command took around 4
seconds.

  Just so you know, sda is the system, and sdb is exclusively for squid cache.

  Does these numbers indicate that I/O operations might be the
bottleneck of the slowdown?

  Thanks,

Felipe Damasio

2010/1/26 John Lauro <john.lauro_at_covenanteyes.com>:
> Yup, your stats on free look fine, especially if squid has been running
> awhile.  If it was recently restarted, it might not be accurate.  Looking
> closer at the data you originally provided, it should be accurate...  You
> may want to add 1-2GB of RAM cache to reduce your disk I/O (if that turns
> out to be the bottleneck).
>
> The vmstat and iostat should eliminate disk I/O as a bottleneck (or point to
> it).
>
>
> You may want to check out the stats with "netstat -s", maybe before and
> after the other commands so you can see the deltas.
>
> Hmmm, I just reread and noticed it was around midnight...  I wonder if a
> bunch of stuff goes invalid in the cache because of the date change?  Never
> noticed that behavior before, but not currently running on a large setup
> either.
>
> Should probably run the vmstat/iostat prior to the extreme slow time to get
> a bit of a baseline for normal operation.
>
>
> Given the high number of connections you have, you may want to consider:
> echo 1024 60999 > /proc/sys/net/ipv4/ip_local_port_range   (probably not an
> issue given transparent mode)
>
> and check to see how close you are coming to your connection tracking limit.
> Probably ok, but could cause connections to require retries if a problem.
>
>
>
>
> PS: You mentioned this is in bridge mode?  I am trying to get squid working
> in bridge mode...  have it working fine in transparent mode as a router, but
> no luck with bridge mode.  Can you send your iptables/ebtables/kernel's
> .config, etc?  I'm going to compile that kernel (2.6.29.6) now, in case it's
> something broke in the more recent kernels.
>
>
>
>> -----Original Message-----
>> From: Felipe W Damasio [mailto:felipewd_at_gmail.com]
>> Sent: Monday, January 25, 2010 10:06 PM
>> To: John Lauro
>> Cc: squid-users_at_squid-cache.org
>> Subject: Re: [squid-users] Squid performance issues
>>
>>   Hi Mr. John,
>>
>> 2010/1/26 John Lauro <john.lauro_at_covenanteyes.com>:
>> > What does the following give:
>> > uname -a
>>
>> uname -a:
>>
>> Linux squid 2.6.29.6 #4 SMP Thu Jan 14 21:00:42 BRST 2010 x86_64
>> Intel(R) Core(TM) i7 CPU @ 9200 @ 2.67GHz GenuineIntel GNU/Linux
>>
>> > While it's being slow, run the following to get some stats:
>> >
>> > vmstat 1 11     ;# Will run for 11 seconds
>> > iostat -dx 11   ;# Will run for 11 seconds, install sysstat if not
>> found
>>
>>   I'll run these tonight.
>>
>> > My first guess is memory swapping, but could be I/O.  The above
>> should help
>> > narrow it down.
>>
>>   I thought that, but actually both top and free -m tells me the same
>> thing:
>>
>>              total       used       free     shared    buffers
>> cached
>> Mem:          7979       5076       2903          0          0
>> 4144
>> -/+ buffers/cache:        931       7047
>> Swap:         3812          0       3811
>>
>>   Swap isn't even touched...even when slow.
>>
>>   But if you think vmstat and iostat can help, I'll run them no
>> problem.
>>
>>   Thanks,
>>
>> Felipe Damasio
>>
>> No virus found in this incoming message.
>> Checked by AVG - www.avg.com
>> Version: 8.5.432 / Virus Database: 271.1.1/2644 - Release Date:
>> 01/25/10 19:36:00
>
>

Received on Wed Jan 27 2010 - 21:55:20 MST

This archive was generated by hypermail 2.2.0 : Thu Jan 28 2010 - 12:00:04 MST