[GE users] Followup project
ppk at ats.ucla.edu
Fri Aug 20 17:13:58 BST 2010
I think what Joshua described is pretty much the same across
many UC campuses. We just install Linux and all open source
software on our 6000+ cores cluster. Any software that has
to be licensed is PIs problem. They have to pay for it from
their grant. We just configure FlexLM license manager so
that only their group can use what they paid for. We do
have few general licenses for Matlab, Mathematica etc.
mainly for occasional general users, but waiting time for
those licenses will be longer or unpredictable.
Pretty much everybody write their own (or from a
collaborator) code in Fortran/C/C++ for serial or using
MPI/OpenMP for parallel. Major expenses are for the nodes
itself, Infiniband interconnect fabric and storage server.
> On Fri, 20 Aug 2010 at 9:38am, andy wrote
>> Our customers are giving us figures which say that hardware,
>> electricity, cooling, administration, OS licenses, DRM system licenses
>> only make up one third of their costs. The remaining money goes into the
>> licenses of the software which runs in the grid. Not talking about the
>> labor costs for the engineers using the Grid. So can the license and
>> support costs of a DRM system have any cost saving potential when you
>> compare it with the value it adds to the bulk of hardware in a data
> I think the operative word in the above is "our customers". That's
> obviously a biased sample. In academia, e.g., the situation is quite
> different. I run a cluster with over 3000 cores (about to expand by
> another 1000 or so) used by many PIs. Our software licensing fees are
> practically zero -- the bulk of the software we run is either open-source
> or in-house. Our support fees are minimal. We don't pay for space,
> power, or cooling (well the PIs do, indirectly, but not in proportion to
> how much we use). The vast bulk of our money goes towards nodes. Why?
> It is simply very, very hard to get the PIs to spend money on anything
> other than cores.
> I'm not at all trying to diminish the value of SGE and it's commercially
> supported model. I'm just pointing out that there are some sizeable users
> for whom the paid version just isn't viable.
To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
More information about the gridengine-users