[Beowulf] Re: building a new cluster

[SC Huang]

--- Jeff Layton <jeffrey.b.layton at lmco.com> wrote:

> SC Huang wrote:
> 
> >Hi,
> >
> >I just posted some timing results from my MPI code here:
> >
> >http://www.geocities.com/schuang21/index.html
> >
> >More are coming soon. It looks like even with gigabit switch (at
> least
> >for this MPI code) using more than 16 nodes is not good...
> >
> >Any comment or suggestion is very welcome. :-)
> >
> Question: Is disk I/O included in these times?


No. There is no I/O (just a few screen output lines). It's the
wall-clock timing of one "time step".


> But if I had to guess, I'd say that Opteron/Myrinet would be
> better than Xeon/Myrinet (looking at the smaller number of
> processes and exptrapolating).


I think so, too. But Myrinet is pretty expensive though (something like
$800 more per node).


> 
> Have you thought about Infiniband? I've heard that IB is
> coming waaay down in price. Plus there are motherboards
> coming with built-in IB (not sure when they'll be out though).


That might be a solution. But we also want to chain our current old
cluster (based on 100 Mbps ethernet switch) into the new cluster. Will
that be a problem (Infiniband + Ethernet)?

SCH


		
_______________________________
Do you Yahoo!?
Win 1 of 4,000 free domain names from Yahoo! Enter now.
http://promotions.yahoo.com/goldrush