Re: Is squid unable to handle the load?

At 13:34 5/26/98 -0400, you wrote:
>Is it clear that the problem is with queueing up for the disk?
>If you don't see this behaviour, it seems to me that there's a likely
>chance that there's some other problem happening.

Well, I don't think I have to restart the process to improve the
performance after removing the redirection. (I usually do restart the
process, just in case).

After reading Markus' response i started to look into disk usage. Here's
what I found:

Under normal load (10-20 clients at a time) which server normally receives
it works fine for weeks. The normal output from 'vmstat' looks like this:

 procs memory page disks faults cpu
 r b w avm fre flt re pi po fr sr s0 s0 in sy cs us sy id
 0 3 0 90308 131924 6 1 1 1 0 1 69 0 254 377 74 1 6 93
 0 3 0 89796 131924 6 1 1 1 0 1 66 0 219 263 71 3 4 93
 0 3 0 89796 131916 7 1 1 1 0 1 62 0 198 241 66 1 2 97
 0 3 0 89796 131900 9 1 1 1 0 1 66 0 211 205 70 2 4 94
 0 3 0 89796 131900 5 1 1 1 0 1 69 0 200 179 73 1 3 96
 0 3 0 89188 131900 5 1 1 1 0 1 70 0 212 208 74 3 4 93

After redirection is enabled for a while, vmstat output looks like this:

 procs memory page disks faults cpu
 r b w avm fre flt re pi po fr sr s0 s0 in sy cs us sy id
 0 3 0 107096 112728 28 1 1 1 0 1 62 0 330 684 67 2 7 91
 0 3 0 106580 112700 12 1 1 1 0 1 63 0 320 660 78 2 6 92
 0 3 0 105928 112700 7 1 1 1 0 1 66 0 335 602 77 5 5 90
 0 3 0 107292 112612 27 1 1 1 0 1 61 0 328 838 71 7 11 82
 0 3 0 107332 112556 19 1 1 1 0 1 61 0 325 728 69 7 8 85
 0 4 0 107816 112500 19 1 1 1 0 1 63 0 294 555 70 4 5 91
 0 3 0 107816 112500 6 2 1 1 0 1 100 0 296 16 10 0 4 96
 0 3 0 107816 112500 5 1 1 1 0 1 112 0 283 12 4 0 0 100
 0 3 0 107816 112500 5 1 1 1 0 1 99 0 274 12 5 0 0 100
 0 3 0 107816 112500 5 1 1 1 0 1 108 0 259 12 4 0 0 100
 1 2 0 107816 112500 5 1 1 1 0 1 110 0 238 14 9 0 0 100
 1 2 0 108100 112376 36 1 1 1 0 1 45 0 245 883 66 28 9 63
 0 3 0 108100 112352 13 1 1 1 0 1 65 0 325 836 71 6 5 89

As you can see when the disk operations / second go above 100 (12th
column), the CPU idle time goes 100% (last column), which basically means
squid freezes...

Correct me if I'm wrong, but that means our SCSI controller (Adaptec 2940
Ultra) / disk is maxing out at 100 operations / second...

Mark Dabrowski

+------------------------------------------<<<
+ Mark Dabrowski @ Canada On-Line @ Toronto, Ontario
+ Tel: (905) 624-8566 ext.5067, Fax: (905) 624-5397, mark@ica.net
+ For technical inquires please email support@ica.net or call our
+ Technical Support Department at (905) 624-1953.
+------------------------------------------------------------------>>>
Received on Tue May 26 1998 - 11:24:29 MDT