0

We are looking at installing Openshift and have the choice of installing on baremetal (64 physical nodes) or Openstack (IaaS consisting of 64 physical nodes).

We are trying to work out the advantages of baremetal vs IaaS (assuming the same hardware profile of both environments)

The main obvious difference seems to be the Openshift node count. Baremetal will always be 1:1, so a maximum of say 60 nodes, however cloud deployments on Openstack can scale Openshift nodes massively as they are just instances.

Does the ability to scale Openshift nodes on Openstack actually matter as you are always limited to the physical resource available underneath anyway, so this should this not be a reason alone for selecting an IaaS Openshift install over baremetal?

Compendius
  • 21
  • 3
  • Virtlualizing gives you so much more flexibility that I'd just do that anyway. Of course you should not have 1:1 virtual-physical machines. This negates many of the advantages! – Michael Hampton Apr 16 '18 at 17:29
  • There are problems with virtualising with Openstack - another layer of complexity to manage, double network encapsulation (means slower networking), more fault domains (baremetal, IaaS (Openstack) and CaaS (Openshift), more complex upgrade paths/dependancies, less visibility of pod to baremetal relationship (more detached) – Compendius Apr 19 '18 at 08:49

1 Answers1

0

After some research the answer is twofold -

  1. Configuration maximums vs resource availability vs pod demand

    (from RH Openshift 3.9)

    • Maximum nodes per cluster 2000
    • Maximum pods per cluster 120,000
    • Maximum pods per node 250
    • Maximum pods per core 10

      If in the unlikely situation you have a small amount of physical nodes and a very high pod requirement it may push you towards a cloud IaaS (Openstack) as the install base to increase the node count.

  2. The management overhead Openstack brings

    If for example you are have the choice of using a third party managed Openstack cloud this may be appealing for the flexibility and API richness this brings to complement Openshift, without any of the Openstack operational fuss. Caveats may be performance compromises based on nested network overlays and virtualisation in general.

    For my situation with 64 physical nodes available the baremetal Openshift deployment will be plenty enough to meet the required node count for our pod requirements and will give maximum performance and operational ease.

Compendius
  • 21
  • 3