[ovirt-users] Spam Cluster Polcies

[Jiri Moskovcak]

On 06/30/2014 03:11 PM, Gilad Chaplik wrote:
> ----- Original Message -----
>> From: "Jiri Moskovcak" <jmoskovc at redhat.com>
>> To: "Gilad Chaplik" <gchaplik at redhat.com>, "Maurice James" <mjames at media-node.com>
>> Cc: "users" <users at ovirt.org>
>> Sent: Monday, June 30, 2014 3:57:29 PM
>> Subject: Re: [ovirt-users] Spam  Cluster Polcies
>>
>> On 06/30/2014 01:53 PM, Gilad Chaplik wrote:
>>> ----- Original Message -----
>>>> From: "Maurice James" <mjames at media-node.com>
>>>> To: "users" <users at ovirt.org>
>>>> Sent: Friday, June 27, 2014 1:20:23 AM
>>>> Subject: [ovirt-users] Spam  Cluster Polcies
>>>>
>>>> Can someone explain the cluster policies to me? The explanation at
>>>> http://www.ovirt.org/Features/Even_VM_Count_Distribution is not quite
>>>> clicking for me.
>>>>
>>>>
>>>>
>>>>       * HighVMSCount - Maximum VM limit. Exceeding it qualifies the host as
>>>>       overloaded. ( Understood )
>>>
>>> minimal #vms on host to enable the module (i.e worst host should have more
>>> than HighVMSCount vms).
>>>
>>>>       * MigrationThreshold - defines a buffer before we start migrating VMs
>>>>       from the host ( Not quite grasping. If my High limit is 5 does this
>>>>       mean
>>>>       that it will only begin to move VMs when the count reaches 10? )
>>>
>>> don't think that there's any relation with "HighLimit".
>>> It means that migration will happen only if there is a difference of X vms
>>> or more between worst host (source host) to destination host.
>>>
>>> let say that you have 2 hosts. one has 10 vms and the second 8. if the
>>> threshold is 3, no migration will happen (10-8 !> 3).
>>
>> - exactly
>>
>>>
>>>>       * SPMVMCountGrace - defines how many slots for VMs should be reserved
>>>>       on
>>>>       SPM hosts (the SPM host should have less load than others, so this
>>>>       variable defines how many VM less it should have) ( Does this mean
>>>>       that
>>>>       if I have a host with only 3 VMs the SPM will 3 VMs minus the
>>>>       SpmWmGrace? Or is this HighVmCount minus SpmVmGrace count?)
>>>
>>> you're right, the SPM host will have current + grace VMs in terms of module
>>> calculations.
>>> I think that this is problematic, it is likely to be selected as worst host
>>> candidate and that's not the wanted behavior iiuc
>>> imo, we should subtract (-) in overloaded hosts and add (+) in
>>> underutilized.
>>
>> - I think it's correct as it is, if the SpmVmGrace+currentVms >
>> HighVmCount it is overutilized and should be freed if possible.
>
> not sure I understand, let's take an example
> 2 hosts: spm: 8 #vms, hsm: 8 #vms.
> SPMVMCountGrace = 2.
> in this case the 'worst' host will be the SPM... shouldn't we deduct 2 from spm and not add? we want less noise there.
>

- that's correct, the SPM will be marked as overutilized and the engine 
will try to find a host to migrate some vms from the SPM. The SpmGrace 
is not about lowering the noise it's about keeping the host less 
utilized by running less VMs on it (which will also lower the noise, 
because it will give the SPM host worse score when looking for target to 
migrate to)

--J

>>
>> --J
>>
>>>
>>> @Jirka?
>>>
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Users mailing list
>>>> Users at ovirt.org
>>>> http://lists.ovirt.org/mailman/listinfo/users
>>>>
>>
>>