high load system do not take all CPU time

[Коньков Евгений]

Здравствуйте, wishmaster.

Вы писали 19 декабря 2011 г., 6:54:08:



w>   --- Original message ---
w>  From: "Коньков Евгений" <kes-kes at yandex.ru>
w>  To: "Daniel Staal" <DStaal at usa.net>
w>   Date: 18 December 2011, 19:47:40
w>  Subject: Re[2]: high load system do not take all CPU time
w>  
w>  


>> Здравствуйте, Daniel.
>> 
>> Вы писали 18 декабря 2011 г., 17:52:00:
>> 
>> DS> --As of December 17, 2011 10:29:42 AM +0200, Коньков Евгений 
>> DS> is alleged to have said:
>> 
>> >> How to debug why system do not use free CPU resouces?
>> >>
>> >> On this pictures you can see that CPU can not exceed 400tics
>> >> http://piccy.info/view3/2368839/c9022754d5fcd64aff04482dd360b5b2/
>> >> http://piccy.info/view3/2368837/a12aeed98681ed10f1a22f5b5edc5abc/
>> >> http://piccy.info/view3/2368836/da6a67703af80eb0ab8088ab8421385c/
>> >>
>> >>
>> >> On these pictures you can see that problems begin with trafic on re0
>> >> when CPU load rise to "maximum"
>> >> http://piccy.info/view3/2368834/512139edc56eea736881affcda490eca/
>> >> http://piccy.info/view3/2368827/d27aead22eff69fd1ec2b6aa15e2cea3/
>> >>
>> >> But there is 25% CPU idle yet at that moment.
>> 
>> DS> <snip>
>> 
>> >># top -SIHP
>> >> last pid: 93050;  load averages:  1.45,  1.41,  1.29
>> >> up 9+16:32:06  10:28:43 237 processes: 5 running, 210 sleeping, 2
>> >> stopped, 20 waiting
>> >> CPU 0:  0.8% user,  0.0% nice,  8.7% system, 17.7% interrupt, 72.8% idle
>> >> CPU 1:  0.0% user,  0.0% nice,  9.1% system, 20.1% interrupt, 70.9% idle
>> >> CPU 2:  0.4% user,  0.0% nice,  9.4% system, 19.7% interrupt, 70.5% idle
>> >> CPU 3:  1.2% user,  0.0% nice,  6.3% system, 22.4% interrupt, 70.1% idle
>> >> Mem: 843M Active, 2476M Inact, 347M Wired, 150M Cache, 112M Buf, 80M Free
>> >> Swap: 4096M Total, 15M Used, 4080M Free
>> 
>> DS> --As for the rest, it is mine.
>> 
>> DS> You are I/O bound; most of your time is spent in interrupts.  The CPU is
>> DS> dealing with things as fast as it can get them, but it has to wait for the
>> DS> disk and/or network card to get them to it.  The CPU is not your problem;
>> DS> if you need more performance, you need to tune the I/O.  (And possibly get
>> DS> better I/O cards, if available.)
>> 
>> DS> Daniel T. Staal
>> 
>> can I get interrupt limit or calculate it before that limit is
>> reached?
>> 
>> interrupt source is internal card:
>> # vmstat -i
>> interrupt                          total       rate
>> irq14: ata0                       349756         78
>> irq16: ehci0                        7427          1
>> irq23: ehci1                       12150          2
>> cpu0:timer                      18268704       4122
>> irq256: re0                     85001260      19178
>> cpu1:timer                      18262192       4120
>> cpu2:timer                      18217064       4110
>> cpu3:timer                      18210509       4108
>> Total                          158329062      35724
>> 
>> Have you any good I/O tuning links to read?
>> 
>> -- 
>> С уважением,
>> Коньков                          mailto:kes-kes at yandex.ru
w>   
w>  Your problem is in the poor performance LAN Card. Guy from
w> Calomel Org told you about it. He advised you to change to Intel Network Card.

see at time 17:20
http://piccy.info/view3/2404329/dd9f28f8ac74d3d2f698ff14c305fe31/

at this point freeradius start to work slow because of no CPU time is
allocated to it or is allocated to little and mpd5 start to drop users because of no response
from radius. I do not know what idle were on 'top', sadly.

does SNMP return right values for CPU usage?

-- 
С уважением,
 Коньков                          mailto:kes-kes at yandex.ru