We have three servers and our plan is to run a number of virtual machines on them in such manner, that if one of the nodes blow up, we can either quickly or seamlessly get a spare running on another node.

In addition to the normal networking, they're interconnected via dual 10Gbit NIC's, so networked raid/mirroring shouldn't be a problem.

The guest VM's are mostly going to be running text mode linux, but of course it wouldn't hurt to be able to spin up a non-mission critical windows guest for running Visual Studio or checking IE compatibility of a web app.

We've spent some time trying to get some magical cloud setup running using Stackops and Crowbar but it started to look like they were offering way too much and were too complicated for our needs.

The next candidate, I think, is Ubuntu 11.04 server + KVM + Ganeti + Drbd, unless you can come up with a suggestion for a better solution that we have missed.


  • Installation should be simple or at least understandable without being in the dev team
  • A browser interface for creating and managing VM's is a nice bonus
  • Single node's hardware failure should cause minimal downtime for VM's that were running on that node
  • Adding more nodes should be possible without shutting down the VM's.

I think this is an easy setup, but that might just be me.

I would install the VMWare Hypervisor on all of the nodes, then on a Windows Server 2008 machine, install the VMWare VSphere Server, and optionally, on the same (or different) server, install the VMware VSphere client (which is what you will manage everything from). Have the nodes setup for HA (high availability), and boom! Done. This will require money for the licenses, but you can at least test it for free under the 30 day eval offered.

Since this is by VMWare, it will obviously have a lot of helpful resources since they are one of the biggest virtual machine software solutions, and have help all over the internet forums.

Under the criteria that you provided, this solution should pass the 1st, 2nd, and possibly the 4th requirement. The 4th requirement can only be met with certain hardware (I think). To manage the VMs (3rd requirement), you will need to use the VMWare vSphere client (special software).

Let me know if you have any questions about this with a comment.

  • Yes, I think VMWare could be a nice solution, but as we don't have any windows workstations, I think we're going to need a linux based setup. There's only a windows client for vSphere and the last time I checked, it doesn't run in Wine either. Also the pricing is quite high, looks like it would be 5k per node as they're all dual cpu. Nov 15 '11 at 19:56
  • 1
    Sorry I couldn't be of more assistance.... I don't know of any others.
    – David
    Nov 15 '11 at 19:59

For a nice fully open source and robust solution I would recommend the Xen Cloud Platform + CloudStack.





To follow on from David's excellent answer (that's how I'd do a full-blown corporate setup if I had the budget!), you can make up an 'economy' version of the setup he describes as follows:

  1. Have a Linux server on the network with sufficient disk space to store a full copy of all your VMs and all their data. Make the storage space on this server available through an NFS share.

  2. Create an NFS datastore on each ESXi server, pointing them to the NFS share.

  3. Use the ghettoVCB script here ( http://communities.vmware.com/docs/DOC-8760 ) to backup the live virtual machines to the NFS datastore. You will have to decide how often you schedule the script to run - and if you run it via a cron job you'll need to check out how to make cron jobs persist across VMware server reboots.

  4. If the primary server goes down, you can fire up the backup copies on the NFS share on another server using the 'add to inventory' menu option.

If you make use of free ESXi licences this is a very low cost setup. The compromises are:

  1. No instant fallover
  2. Manual intervention required when an outage occurs

I have tested this setup in-house and it works well for us.

  • A crashing node means data loss between the last sync and the crash. Also, ESXi isn't free on multi-cpu hosts. But I still like the ghetto tech approach :) Nov 15 '11 at 23:13
  • Oh yes, it's not a perfect solution by a long way - as I said; it's the 'economy' version!
    – Linker3000
    Nov 16 '11 at 12:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.