[Beowulf] Using commercial clouds for HPC

[stephen mulcahy]

Bill Broadley wrote:
> I built a small 16 node cluster on EC2, and it worked well.  I configured the
> cluster to use MPI, a shared file system via nfs, and a batch queue.  This was
> before their locality API was available so I had to do some snooping about to
> figure out how close my nodes were.  It ended up performing much like a normal
> GigE connected cluster, suitable for any job that wasn't too interconnect
> intensive.  Overhead in virtualization for the CPU and memory system was
> rather small, seemed under 20% which isn't too big a deal since you can always
> just allocate more nodes.

Hi Bill,

Thanks for the feedback. Your results seem to agree with the paper 
Rayson referred to in his mail. It looks like for hpc jobs that happily 
run on gigabit, EC2 is a reasonable option (with around a 20% overhead 
from virtualisation). It doesn't look like EC2 is currently suitable for 
hpc jobs that require infiniband type interconnects to run in a 
reasonable timeframe. But as you suggest, it may make sense to purchase 
a few hours of compute time and play around.

Thanks,

-stephen

-- 
Stephen Mulcahy     Atlantic Linux         http://www.atlanticlinux.ie
Registered in Ireland, no. 376591 (144 Ros Caoin, Roscam, Galway)