[Users] oVirt warning about memory usage

Amador Pahim apahim at redhat.com
Tue Jan 8 17:34:17 UTC 2013


On 01/08/2013 11:38 AM, Alex Leonhardt wrote:
> thanks, yep noticed it has a negative value -
>
> Amador, does that mean the host is overcommitting by 18 G of avail 
> physical memory although 20 G of phys memory is still free ?

memAvailable maps the worst case to the Host.
It means: considering your current state, to assure the "Physical Memory 
Guaranteed" for all running VMs, you need 18GB of additional RAM. This 
is the consequence of enabling overcommit policy.

memAvailable = FREE_OR_CACHED + RESIDENT - memCommitted - Host Reserved 
Memory

Where:
  FREE_OR_CACHED: from /proc/meminfo
  RESIDENT: RSS value (2nd. field in /proc/<PID>/statm) from all running 
VMs processes.
  memCommitted   -->  Physical Memory Guaranteed + Guest Overhead 
(default 65MB) from running VMs.

*Notice that memAvailable does not considers KSM work.

>
> Alex
>
>
>
> 2013/1/7 Amador Pahim <apahim at redhat.com <mailto:apahim at redhat.com>>
>
>     On 01/07/2013 06:18 AM, Alex Leonhardt wrote:
>>     Hi,
>>
>>     just seen a memory usage warning in oVirt's Admin Interface
>>     saying the available memory is below threshold of 1024MB -
>>     however - when I checked the host, it had still 20GB left ?
>>
>>     See screenshot.
>>
>>     Alex
>>
>>
>>
>>     _______________________________________________
>>     Users mailing list
>>     Users at ovirt.org  <mailto:Users at ovirt.org>
>>     http://lists.ovirt.org/mailman/listinfo/users
>
>     The message is happening due to "memAvailable" statistic coming from vdsm. The memAvailable field can be negative and considers a set of values from different sources to compose its result (free memory, cached memory, resident set size from qemu processes, committed memory from running vms and host reserved memory
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20130108/e586d9a3/attachment-0001.html>


More information about the Users mailing list