[one-users] Scheduler Policy set to Custom for More enough Memory and Less CPU load

[Ruben S. Montero]

Hi

When using a placement heuristic that combine different metrics, you
will need to:

1.- Scale all the metrics so you can "mix" them. Usually this is done
by scaling the metric to 1. For example, if the max memory and cpu in
your cluster are 4096 and 800, respectively. You will need to divide
FREEMEMORY by 4096 and FREECPU by 800.

For example a host with FREEMEMORY=2048 and FREECPU=720 will give you
scaled values of 0.5 and 0.9, respectively.

2.- Linear combination of the policies. You basically have two
policies to sort the hosts, those with more FREEMEMORY and those with
more FREECPU. What is more important?. This can be done by multiplying
the scaled value by a constant

In the above example, I'd set the placement heuristic as:

 A * (FREEMEMORY/4096) + B * (FREECPU/800)

If both attributes are equally important then A=B=1, so

DEFAULT_SCHED = [
    policy = 3,
    rank   = "(FREEMEMORY/4096) + (FREECPU/800)"
 ]


Hope it helps


Cheers

Ruben

On Fri, Oct 5, 2012 at 12:00 AM, Lawrence Chiong <junix88 at gmail.com> wrote:
> Hello,
>
> Could anyone correct my scheduler configuration  (/etc/one/sched.conf) using
> custom policy. Supposing I want to prefer hosts to deploy VM with more spare
> of Memory and less CPU load. Defined below -
>
> DEFAULT_SCHED = [
>     policy = 3,
>    rank   = "(FREEMEMORY - FREECPU * 100)"
> ]
>
> Any help and suggestions is very much appreciated.
>
> Thank you.
>
> Regards,
>
> junix
>
>
> _______________________________________________
> Users mailing list
> Users at lists.opennebula.org
> http://lists.opennebula.org/listinfo.cgi/users-opennebula.org
>



-- 
Ruben S. Montero, PhD
Project co-Lead and Chief Architect
OpenNebula - The Open Source Solution for Data Center Virtualization
www.OpenNebula.org | rsmontero at opennebula.org | @OpenNebula