1

I just setup virtual machines on one hardware with Vagrant. I want to use a Puppet to configure them and next try to setup OpenStack. I am not sure If I am understanding how this should look at the end. Is it possible to have below architecture with OpenStack after all where I will run one Virtual Machine with Linux?

-------------------------------
|          VM with OS         |
-------------------------------
| NOVA    | NOVA    | NOVA    |
-------------------------------
|          OpenStack          |
-------------------------------
|  Node   | Node    | Node    |
-------------------------------

More details:

In my environment Nodes are just virtual machines, but my question concerns separate Hardware nodes. If we imagine this Nodes(Novas) are placed on a separate machines (e.g. every has 4 cores) can I run one Virtual Machine across many OpenStack Nodes? Is it possible to aggregate the computation power of OpenStack in one virtual distributed operating system?

Knight of Ni
  • 137
  • 1
  • 6
  • You drew a picture, but perhaps you should explain it instead. It's not really clear exactly what you want to do. – Michael Hampton Nov 04 '13 at 00:37
  • I think your understanding of the technologies and techniques involved here need work. I suggest you try to understand your problem more - why do you want a VM with 12 cores on a host with only 8? Why have you selected OpenStack? Why are you trying to use virtualisation at all? – Craig Watson Nov 04 '13 at 13:15

1 Answers1

6

Can I run one Virtual Machine across many Nodes with OpenStack? Is it possible to aggregate the computation power of separate machines in one virtual distributed operating system?

No, you've got OpenStack by exactly the wrong end. VM's can only run on one single bare iron node. The whole point of OpenStack is generally to create more VM's than you have bare iron nodes and to make putting up/taking down a VM easy.

Most attempts to "aggreate computation power" are focused on the application level. Depending on exactly what you need to transfer between nodes there are many solutions out there. OpenMPI is fairly standard in the high performance computing world, but it's a message passing library. Other approachs like Hadoop attempt to bring the computation to the data.

A "virtual distributed operating system" is certainly possible and many attempts have been made over the years. I don't know of any that ever made it out of the experiment stage though.

"Plan 9" comes reasonably close to matching your description.

http://en.wikipedia.org/wiki/Plan_9_from_Bell_Labs

Fred the Magic Wonder Dog
  • 2,095
  • 12
  • 17
  • TomTom usually references a vendor that will allow you to span a VM across multiple physical servers. Performance not guaranteed, of course. – mfinni Nov 05 '13 at 17:15
  • ScaleMP is the vendor, had to dig around for that. – mfinni Nov 05 '13 at 22:48