Bug #1007

Host finding for VM

Added by Max Mikheev over 9 years ago. Updated about 8 years ago.

Status:ClosedStart date:11/24/2011
Priority:NormalDue date:
Assignee:-% Done:

0%

Category:-
Target version:-
Resolution:worksforme Pull request:
Affected Versions:OpenNebula 3.2

Description

Hi,
OpenNebula has great feature to find host which satisfy to required resources for VM but it is not work every-time well.
Case:
I have 12 core machine and would like to create 2 instances with up to 8 cores (800%) on it. I know that this two machines will never completely load together. Both of them has peak performance with full load less than 10% of time. When machine is not loaded it use 1-3% of processor only. I don't care if both machines will be activated in same time and will get 600% instead of 800% during a peak. I am sure that I would like to have both machines running on a same host.
1. I created one VM instance.
2. I am trying to create a second one. If a first machine is not loaded the second can be created without any problem.
If a first one is loaded than the second one could not be created. I received message in the sched.log: " [SCHED][D]: Host 0 filtered out. It does not have enough capacity."

Looks like OpenNebula checked CPU amount for a new VMs and compared it with FREECPU for a host. If cpu's for VMs value larger than FREECPU host is rejected.

The suggestion is to add minimal amount of CPU for a VM which will be compared to FREECPU instead of comparing maximal CPU with FREECPU. In many cases guest never use maximal amount of CPU and we are wasting unused processors power.

Thanks

History

#1 Updated by Ruben S. Montero over 9 years ago

  • Target version deleted (Release 3.4)

#2 Updated by Ruben S. Montero about 8 years ago

  • Status changed from New to Closed
  • Resolution set to worksforme

the CPU parameter can be used to overcommit and the cgroups/xen credits can manage the peaks.

Also available in: Atom PDF