[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Huge AMD Opteron cluster installed with FAI and Debian amd64


> > Btw. a nice way to increase the size of your cluster is simply to ask
> > the technicians for old hardware (which means everything >600MHz) that
> > is no longer used. Usually, these computers are from student labs and
> > would otherwise go to charity (which is of course fine as well...).
> > 
> Am particularly interested how you make use of a non-uniform cluster.
> Most of the applications (MPI) we've developed will run as fast as the
> slowest node.

We're using PBS (tried both OpenPBS and PBSpro) - there the node
attributes are useful to define different groups of nodes (we're
basically using slow/fast and low/high memory). I'm usually using the
queuing mode, here of course every additional node is fine. But for
applications using MPI it usually came down to arrangements with
colleagues as to who gets which set of nodes anyway. Therefore our
approach always was to include whatever we got...


Reply to: