We are studying implementing some virtualized servers here, but we don't know what will be better suitable for us. Some folks are saying better have two huge servers, and others are saying have like a ten middle-end servers.

We have a legacy Visual Foxpro application, which nowadays run on Dual Xeon E5405 @ 2GHz and 16Gb of RAM. The currently server its getting too slow due the number of active users and process running on it. Virtualizing this server will give us the benefit of an faster disaster recovery.

So the question is, having like ten physical servers running at 1.7GHz and 4Gb of RAM, we could virtualize one server into 4 machines, and have one virtualized server running at 6.8GHz and 16Gb of memory? If yes, there is some how if one machine stops, automatically manage this virtual machine to another one, and execute the appropriate maintenance on it, and later back to it again?

  • 2
    FoxPro? Wow. Could you not have one, for example, HP Server with dual sept-core CPUs and a ton of RAM instead of a lot of machines clustered togther? You power bill would likely be less, if anything else.
    – tombull89
    Oct 18, 2012 at 14:56
  • The answer for the question in your third paragraph is in most circumstances NO, it does not work like that. You cannot normally take 4 1.7Ghz 4GB machines and use virtualization to create a single a 6.8Ghz 16GB machine. However, if you have a single server with two 3.4Ghz processors and 16GB of RAM, you can easily run 4 separate instances of Windows Server on that machine.
    – Joel Coel
    Oct 18, 2012 at 16:14
  • 1
    I think that this is going to be the next phase in virtualization. I could care less about putting more virtual machines on a physical machine.. I'd LOVE to be able to take 10 desktops and combine their resources into a single Virtual Server. Oct 18, 2012 at 18:39
  • it is not the next Phase, it already there. For - hm - more than 50 years. Mainframes pretty much work like that. And today there is one Company already offering it.
    – TomTom
    Oct 18, 2012 at 18:51
  • 1
    so I can run one instance of Windows across 20 mainframes? I didn't think so! Oct 18, 2012 at 21:23

2 Answers 2


Yes, you can combine multiple x86 machines into a larger virtual x86 machine, with ScaleMP. Compatible with Xen and KVM Hypervisors, you can then create VMs that will span multiple physical machines.

You could then run a large windows VM within your Xen or KVM hypervisor on top of your ScaleMP cluster.

Here's a write up that's a bit easier to read than their website: http://www.readwriteweb.com/solution-series/2011/10/cost-effective-clustering-with.php

  • 3
    +1 for being one of the fhew that know what they talk about - this is definitely not mainstream tech, but it IS possible. This pops up regualrly and most of the time most answers just point out how clueless people can be - they simply do not know what is possible.
    – TomTom
    Oct 18, 2012 at 15:51
  • 1
    While technically correct, this would be totally inappropriate for what the OP is trying to accomplish, at least based on the information available.
    – Joel Coel
    Oct 18, 2012 at 16:18
  • 2
    how so? OP requested a way to have one virtual server span multiple physical machines. also, he mentioned a legacy Virtual FoxPro application, which mean he'd need to run Windows as the Virtual Server OS. Running Windows, in Xen or KVM, on top of ScaleMP would EXACTLY enable what the OP was requesting. Oct 18, 2012 at 16:33
  • This is awesome! I had no idea this existed. I was assuming the answer is "no, this flat out isn't possible because performance would be terrible"
    – Earlz
    Oct 18, 2012 at 19:50
  • 2
    @Earlz "Possible" and "Bad Performance" are two rather different things, though ;)
    – Izkata
    Oct 18, 2012 at 20:54

You have a fundamental misunderstanding of virtualization. You can not take one virtual machine and spread it across multiple virtualization hosts. Virtualization lets you run multiple virtual machines on a single host.

enter image description here

  • 2
    To be more precise: disks can be combined in this way (it makes sense, as network speed usually greatly exceeds disk access speed), while memory and CPUs cannot (the resulting virtual memory and virtual CPU would be too slow). Oct 18, 2012 at 14:42
  • 16
    -1. Learn before talking. See, yo uare right regarding MOST commercial level hypervisors. Seriously. But this is not how it HAS to work, and yes, the other side is possible - there is at least one vendor out that does that. Takes serious hardware, though (infiniband to handle the memory bandwidth), comes with overhead, and hte price is steep, but it basically simulates a NUMA machine where every physical machine is a NUMA node. Google is your friend ;) Yes, there is one. Or you read the correct answer by Tim Kennedy ;)
    – TomTom
    Oct 18, 2012 at 15:50

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.