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