[openstack-dev] [nova][baremetal] Scheduling baremetal deployment on different hw model

Taurus Cheung Taurus.Cheung at harmonicinc.com
Mon Jun 30 05:18:56 UTC 2014


Hi Rob,

Thanks your reply.

As I know, Ironic not yet "graduate". It is still under rapid development as replied by Chris Krelle in Feb-2014: http://lists.openstack.org/pipermail/openstack-dev/2014-February/026647.html . Or I should change to Ironic now?

No - I am not group IBM and Dell machines together. They are under the same datacenter and I want to provision them under a single OpenStack controller, which I think a very common use case. What I want to do is to choose specified hw model (IBM or Dell) when I deploy image in baremetal style.

Regards,
Taurus

-----Original Message-----
From: Robert Collins [mailto:robertc at robertcollins.net] 
Sent: Monday, June 30, 2014 12:46 PM
To: OpenStack Development Mailing List (not for usage questions)
Cc: Taurus Cheung
Subject: Re: [openstack-dev] [nova][baremetal] Scheduling baremetal deployment on different hw model

Firstly, use Ironic. Nova BM is deprecated.

Secondly, yes, you can use extra-specs, but you only need to do that if your machines are identical in CPU, disk and memory - which the scheduler will look at anyway. Why do you need to group IBM and Dell machines together?



-Rob

On 30 June 2014 16:26, Taurus Cheung <Taurus.Cheung at harmonicinc.com> wrote:
> Hi,
>
>
>
> I am working on deploying images to bare-metal machines using nova 
> bare-metal. My datacenter has 2 types of hw models, IBM and Dell. In 
> existing implementation, if I want to deploy image on specified type 
> of hw model, I need to setup 2 baremetal compute nodes, one for 
> container of IBM machine, the other for Dell machine. Then baremetal 
> register machines to their corresponding compute node. Finally use 
> nova flavor and heterogeneous group to map specified compute node so I 
> can explicitly specify the hw model to deploy, as illustrated as following flow chart:
>
>
>
> Flavor_IBM -> (mapping by flavor extra_spec) -> 
> Heterogeneous_Group_IBM -> Compute_Node_IBM -> IBM machines
>
> Flavor_Dell -> (mapping by flavor extra_spec) -> 
> Heterogeneous_Group_Dell -> Compute_Node_Dell -> Dell machines
>
>
>
> The existing approach has a drawback: I need to setup 1 baremetal 
> compute node for each hw model. If I have 10 hw models in my 
> datacenter, I need to setup 10 baremetal compute node, which would be 
> a high overhead. Is there any update in ironic to tackle this?
>
>
>
> I think one of the possible enhancement is adding a field like 
> hw_model in nova.bm_nodes DB and passing to nova scheduler, so 
> different hw models of machine can under the same baremetal compute node and heterogeneous group.
> Just using different extra_spec in nova flavor to specifiy hw_model. 
> Is it a good idea?
>
>
>
> Regards,
>
> Taurus
>
>
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>



--
Robert Collins <rbtcollins at hp.com>
Distinguished Technologist
HP Converged Cloud


More information about the OpenStack-dev mailing list