Re: [squid-users] Queue congestion at 60 req/sec

On Sun, Jun 10, 2007, Pablo Garc?a wrote:
> The objects being cached are 32K at most,
> this is the busiest line of iostat -x I got
>
> avg-cpu: %user %nice %sys %iowait %idle
> 2.49 0.00 1.99 22.39 73.13
>
> Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s rkB/s wkB/s
> avgrq-sz avgqu-sz await svctm %util
> ida/c0d0 0.00 35.64 14.85 3.96 182.18 1148.51 91.09 574.26
> 70.74 3.89 16.32 8.68 16.34
> ida/c0d1 0.00 58.42 36.63 10.89 435.64 570.30 217.82 285.15
> 21.17 0.81 16.56 6.98 33.17

Thats pretty busy; look at the writes. Writing 32kbyte objects to disk
is ~2mbytes/sec write, assuming 100% cachability rate.

See if you can figure out what's using up the disk IO on that disk(s).
The other alternative is to use COSS for the small objects..

Adrian

>
> This is a partial run for vmstat
>
> vmstat 1
> procs -----------memory---------- ---swap-- -----io---- --system--
> ----cpu----
> r b swpd free buff cache si so bi bo in cs us sy id
> wa
> 1 0 208 14528 105372 293208 0 0 75 58 0 7 1 1 88
> 10
> 0 1 208 15168 105288 292512 0 0 400 276 1508 715 2 1 74
> 22
> 0 0 208 14784 105324 292996 0 0 392 0 1541 932 3 1 79
> 18
> 0 0 208 15296 105200 292600 0 0 580 0 1694 1408 3 3 63
> 30
> 0 0 208 14912 105252 292808 0 0 304 0 1598 869 3 2 77
> 19
> 0 2 208 14592 105332 292988 0 0 400 0 1538 971 3 1 70
> 26
> 0 0 208 15552 105284 291996 0 0 348 972 1658 970 3 2 61
> 34
> 0 0 208 14976 105340 292720 0 0 568 0 1710 1158 4 2 68
> 26
> 0 2 208 14528 105400 292920 0 0 408 0 1615 965 3 1 75
> 22
> 0 0 208 14976 105252 292808 0 0 524 0 1644 1074 4 2 70
> 24
> 0 2 208 14592 105296 293024 0 0 376 0 1627 999 3 2 73
> 23
>
>
> It doesn't seems so busy, any ideas ?
>
> Best Regards, Pablo
>
> >
> >
> >
> >
> >On 6/8/07, Adrian Chadd <adrian@creative.net.au> wrote:
> >> On Fri, Jun 08, 2007, Pablo Garc?a wrote:
> >> > Hi, is there any scenario, where I can find this message in the
> >> > cache.log for that amount of requests ? What I'm doing wrong ?
> >> > squidaio_queue_request: WARNING - Queue congestion
> >> >
> >> >
> >> > Running as reverse proxy
> >> > 2 scsi disks with cache_dir configured at 60% of the size
> >> > ext3 mounted with noatime
> >> > REDHAT kernel 2.6.9-55.ELsmp (4 Update 5)
> >> > 2 x Pentium III 1Ghz. 1Gb RAM
> >> > access_log none
> >> > cache_store_log none
> >> > cache_mem 256 Mb
> >> >
> >> >
> >> > CPU at 16% IOWAIT, 4% USER and 2,6% SYSTEM
> >> > 60 client req/sec, 70% hit Ratio.
> >>
> >> Well, you haven't given the one actual bit of info you need, which is IO
> >> throughput. Run "vmstat 1" and "iostat 1", look at the device IO.
> >> Are your objects big or small? If they're big then each object is going
> >> to generate a whole lot more IO than smaller objects, hence the queue
> >> congestion.
> >>
> >>
> >>
> >> ADrian
> >>
> >>
> >

-- 
- Xenion - http://www.xenion.com.au/ - VPS Hosting - Commercial Squid Support -
- $25/pm entry-level bandwidth-capped VPSes available in WA -