[Beowulf] Re: Cluster newbie, power recommendations
Joe Landman
landman at scalableinformatics.com
Tue Mar 21 10:06:17 PST 2006
David Mathog wrote:
> Interesting question that, how else might this be done?
>
> I wonder if one couldn't set up a single modern computer,
> with a fast CPU and tons of memory, as N virtual machines, for
> instance using VMware, and then run a sort of virtual cluster?
Ouch... VMWare is really nice, but the VM slows things *way* down.
> Obviously there wouldn't be any performance advantage to doing
> this but it might allow the subject to be taught when a real
> clusters isn't available.
Not sure of the performance impact of this, but you could look at OpenVZ
or Xen as well (when it is ready). Something like this is probably the
way folks may go anyway so you can boot a cluster of any flavor from a
good base load (OpenVZ'ed Warewulf anyone?)
> Presumably VMware already has some method for handling the network
> needs of each virtual machine, although I'm guessing it might
Yes it does.
> not extend all the way to a DHCP boot emulation. There would also
It can in theory do that as well. I stopped using my copy of VMware in
the 4.0 days. It is up to 5.5, so it might be quite a bit better. I
went back to dual boot because of the speed.
> need to be some sort of peculiar performance measurement system
> at the VMware level since it would have to pretend that the
> N machines running sequentially were actually working in parallel,
> and report performance as if they had been.
Would be hard. Should just get a dual dual-core machine and run a
virtual process per core. Gives you a cluster of 4 nodes.
Of course, why do that when you have this beautiful 4-way SMP...
Sometimes it is better to "just say no" to distributed computing, and
this is one of the cases :)
> Seems doable. Maybe even worth doing so that this subject could
> be more easily taught.
I find that students have no trouble launching 4-8 threads on a single
node to test code, and then have them run it on the real cluster/SMP
using MPI. For OpenMP they are more hindered by the lack of free tools
(there are multiple free MPI tools, no free OpenMP tools that I am aware
of... discounting the "free" Intel compilers which cannot be installed
on a teaching cluster due to the license which prevents *paid* teachers
from installing it for free on a teaching cluster).
> Regards,
Joe (teaching a course on MPI/OpenMP application programming right now)
--
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax : +1 734 786 8452
cell : +1 734 612 4615
More information about the Beowulf
mailing list