Ask Your Question

How many nodes is needed to run distributed OpenStack

asked 2018-09-18 00:36:58 -0500

soheil08 gravatar image

I've read the OpenStack documents and I found that at least two node (controller and compute) is mandatory for running OpenStack in the production environment, but what I observe is another network node is needed, too. According to my understandings in OpenStack architecture controller's duty is the management of instances created on compute nodes in addition to running database and message broker. The compute node will run the hypervisor. So my questions are the following:

  1. If I want to run Big Data VMs (VMs are created in VirtualBox), What is the architecture of the OpenStack nodes? For example, if we have 8 physical servers, I guess we should install controller and network node in one server (one virtual machine for each) and install compute node on the rest of servers (7 servers). Is it a good plan?
  2. If the network node is not mandatory, in which node should I install network component?
  3. Does OpenStack Support High Availability and Fault-tolerant? For example, what happens if the network component or other main components failed?

I checked Sahara component for Big Data and I guess it is useful when we want to create Big Data cluster from scratch, but in my case I already have VMs.


edit retag flag offensive close merge delete

1 answer

Sort by » oldest newest most voted

answered 2018-09-18 05:07:00 -0500

You don’t have to separate the network and general control components. I doubt that eight compute nodes cause enough traffic that a separate computer to process networking is warranted. I would deploy all control services on a physical server, or 2 or 3 physical servers for availability. Not in virtual servers.

I don’t understand what you want to do with Virtualbox. It’s not suitable for production, and it’s not supported by OpenStack.

Yes, almost all OpenStack components are ready for HA. They are stateless, so that you can easily front them with a load balancer. The only difference used to be Cinder Volume, which can’t run in an active/active configuration (perhaps that’s changed), and nova-consoleauth, which can only run once, but it’s deprecated in Rocky anyway. The message queue infrastructure has HA built-in, and for the database you need to select a solution like Percona.

edit flag offensive delete link more


I already created my VMs using VirtualBox. You mean I can't import them to OpenStack even if I export them to ova or ovf format?

soheil08 gravatar imagesoheil08 ( 2018-09-18 05:15:52 -0500 )edit

That probably works. OpenStack supports hypervisors like KVM, Xen, or Hyper-V. There is a short paragraph in the documentation, and Google can provide more detail.

Bernd Bausch gravatar imageBernd Bausch ( 2018-09-18 05:55:26 -0500 )edit

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower


Asked: 2018-09-18 00:36:58 -0500

Seen: 561 times

Last updated: Sep 18 '18