OpenVZ Forum: Support » *SOLVED* Question about cpuunits comment on swsoft forum

Home » General » Support » *SOLVED* Question about cpuunits comment on swsoft forum

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

*SOLVED* Question about cpuunits comment on swsoft forum [message #9410]

Sun, 31 December 2006 17:22

rickb
Messages: 368
Registered: October 2006

Senior Member

Hi, I found this post on the swsoft virtuozzo forum- OVZ and VZ share the same architecture so I feel the question/answer is relevant to openvz as well. I see the person who offered an answer is from Herndon, VA so I am assuming he works for swsoft.

Is his answer about cpuunits correct?

http://forums.sw-soft.com/showthread.php?s=05fab1a5a307d51be 4c6ae9d05bb260a&threadid=38376

The ability for a VE to saturate the disk i/o is a problem I wrestly with daily, so I am interested in what this guy is saying. Basically he is saying a low cpuunits could cause the disk performance to suffer. I'm not sure if this is true because if a VE is not being selected by the scheduler very often (condition from low cpuunits), would the disk be "waiting" or "busy" as seen by other VEs? Its my understanding that a VE can use its cputime to hammer the disk all it wants, but if the cpuunits is very low like in the post, the VE would have no time to hammer the disk and thus effect other VEs.

Long winded, I know. But, looking for some insight.

Rick

-------------
Common Terms I post with: http://wiki.openvz.org/Category:Definitions

UBC. Learn it, love it, live it: http://wiki.openvz.org/Proc/user_beancounters

[Updated on: Tue, 09 January 2007 06:25] by Moderator

Report message to a moderator

Re: Question about cpuunits comment on swsoft forum [message #9443 is a reply to message #9410]

Wed, 03 January 2007 12:22

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

Rick, you misinterpreted Barmaley's words imho.

The issue is very simple: when you limit some VE to very low CPU rate, its processes begin to run slowly, so it takes more time for them to handle a single request or do some other activity. From kernel point of view, such processes though are not given CPU time are still being in RUNNING state (i.e. ready to run, though they are delayed by fairscheduler due to low VE rate). So as the result of the low rate, the number of tasks in VE in RUNNING state is increased. This results in increasing of load-average value shown by `uptime` tool, since it is calculated as an average from (number of running+uninterruptible tasks).

So due to low CPULIMIT loadavg can go higher. But this has nothing to do with I/O.