[one-users] CPU and Memory Overcommitment

[Carlos Martín Sánchez]

Hi,

On Wed, Sep 4, 2013 at 12:03 PM, Igor Laskovy <igor.laskovy at gmail.com>
 wrote:
>
> One further question - how Fault Tolerance mechanism (via HOST_HOOK) deal
> with this reservations? Does it reserved some "slots" for VMs recovery?
>

OpenNebula does not have any kind of reservation scheduling.


> If not and if I at every turn manually don't control available resources
> on hosts those may cause to situation when my automatically recreated VMs
> will stuck in placement state.
>

Yes, that may happen. But you still can implement a reservation mechanism.
I'm sure there are better alternatives, but this is a quick idea:

- Disable half of your hosts.
- For each enabled host, define a failover_host = <id> attribute in its
template, pointing to one of the disabled hosts
- Modify the fault tolerance hook to make it enable the defined failover
host.

Another quick hack is to create a dummy VM for each VM that you need to
guarantee that reserved slot, using the requirements and current_vms
features to deploy it in a host different from the original VM. Then delete
that VM when the hook recreates the original VM...


Regards.

--
Join us at OpenNebulaConf2013 <http://opennebulaconf.com> in Berlin, 24-26
September, 2013
--
Carlos Martín, MSc
Project Engineer
OpenNebula - The Open-source Solution for Data Center Virtualization
www.OpenNebula.org | cmartin at opennebula.org |
@OpenNebula<http://twitter.com/opennebula><cmartin at opennebula.org>


On Wed, Sep 4, 2013 at 12:03 PM, Igor Laskovy <igor.laskovy at gmail.com>wrote:

> Thank you for reply!
>
> One further question - how Fault Tolerance mechanism (via HOST_HOOK) deal
> with this reservations? Does it reserved some "slots" for VMs recovery? If
> not and if I at every turn manually don't control available resources on
> hosts those may cause to situation when my automatically recreated VMs will
> stuck in placement state.
>
>
> On Wed, Sep 4, 2013 at 12:23 PM, Carlos Martín Sánchez <
> cmartin at opennebula.org> wrote:
>
>> Hi,
>>
>> On Tue, Sep 3, 2013 at 1:32 PM, Igor Laskovy <igor.laskovy at gmail.com>wrote:
>>
>>> Hello all!
>>>
>>> I found that this already have discussed not so far from now -
>>> http://comments.gmane.org/gmane.comp.distributed.opennebula.user/10568
>>>
>>> As understand for now about Memory over-commitment for production I can
>>> forget ;)
>>>
>>> For CPU over-commitment I can only use CPU & VCPU attributes, right?
>>>
>>
>> That's right
>>
>>
>>>  If I will, for example, set CPU to 0.2, than host will do only
>>> reservation of processor time for that VM OR do limit ether, so this VM
>>> will limited of 1/5 of one physical/logical hardware core?
>>>
>>
>> We enforce the CPU reserved at hypervisor level with cgroups for kvm,
>> credit scheduler for xen, and the esx cpu scheduler for vmware
>>
>> Regards
>>
>> --
>> Join us at OpenNebulaConf2013 <http://opennebulaconf.com/> in Berlin,
>> 24-26 September, 2013
>> --
>> Carlos Martín, MSc
>> Project Engineer
>> OpenNebula - The Open-source Solution for Data Center Virtualization
>> www.OpenNebula.org <http://www.opennebula.org/> | cmartin at opennebula.org
>>  | @OpenNebula <http://twitter.com/opennebula>
>>
>
>
>
> --
> Igor Laskovy
> facebook.com/igor.laskovy
> studiogrizzly.com
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opennebula.org/pipermail/users-opennebula.org/attachments/20130905/80de819e/attachment-0002.htm>