[vbox-dev] VirtualBox on cluster ( not the cluster on VirtualBox!) - parallelization.

Nikolay Igotti Nikolay.Igotti at Sun.COM
Tue Nov 17 13:46:41 GMT 2009


    Hi Victor,

 No, current VirtualBox doesn't support such mode of operations, and 
unlikely will. Although it theory, with very fast interconnects it would 
be possible, practically such solution will have horrible performance 
and rather dubious advantages. Usually, some form of load balancing with 
minimal data sharing (such as remote execution of hard computations, 
with shared data on iSCSI RAM disk) will provide better use of 
computational resources with traditional hardware, compared to such a 
cluster.
 
 Some big NUMA machines, technically is exactly such as cluster, 
RoadRunner http://en.wikipedia.org/wiki/IBM_Roadrunner being an extreme 
case, but usually OS must be aware about non-uniform nature of 
underlying hardware and behave accordingly.

  Overall, it looks more like a research matter, not smth to implement 
in production virtualizer.

   Nikolay


Victor Vasilchenko wrote:
> Good day,
>
> is it possible to run* VirtualBox on the cluster* (SSI - single-system 
> image) with *the only purpose - to create a single virtual machine 
> with all available* CPUs(cores) and memory ?
>
> Example:
> the set of 4 PCs (CoreQuad + 8GB RAM) forms a Linux cluster with 16 
> cores and 32GB memory.
> Would it be possible to run VirtualBox on this cluster - that is to 
> run it parallelized on all 4 cluster nodes (16 cores),
> the result would be a VirtualPC with 15Cores and 30GB Memory (-1 for 
> host itself).
>
> Just imagine that any OS installed inside this virtual machine will 
> see 15 cores and 31GB memory !
>
> And all the current modern programs will fully utilize it's power 
> (simply multithreading - no need to redesign for MPI support):
>
> 1) 3D studios (Maya, Max) will run on 15 threads
> 2) Video transcoding software will use 15 threads
> 3) make -j 15
> 4) etc....
>
> the only question here is whether VirtualBox can be run on cluster in 
> parallelized manner, but not on just one node of it.
>
> Did you have any experiments for it or may be have any ideas?
>
> Thanks and Regards,
> Victor.
> ------------------------------------------------------------------------
>
> _______________________________________________
> vbox-dev mailing list
> vbox-dev at virtualbox.org
> http://vbox.innotek.de/mailman/listinfo/vbox-dev
>   





More information about the vbox-dev mailing list