[Beowulf] torus versus (fat) tree topologies

Chris Sideroff cnsidero at syr.edu
Mon Nov 15 19:01:58 PST 2004

Mark Hahn wrote:
> that's quite reasonable if you're planning to use off-the-shelf hardware.
> that is, if you're an engineer doing research using HPC.  if you're 
> actually doing research into the implementation of CFD using HPC,
> then you should probably look a bit closer at adaptivity, for instance,
> which winds up making FEM much less nearest-neighbor...

   The former but my personal interests lie in the latter.

> the port count indicates that's an older-generation switch, probably
> with poorer bandwidth than current models.

   It handles up to 48 but there is only 6 (8 max) 6-port line cards 

> I'd be astonished if all of the tier-1 vendors didn't have a test cluster
> available for your asking, probably with fluent installed.

   Maybe so be none have come forward so far other than Dolphin.  I 
might have the opportunity to test some quadrics hardware ...

> there's always a danger of over-benchmarking, but you should probably
> see if you can get access to an IB cluster.  for CFM, I'm a little 
> surprised you appear to care so much about latency, since I'd expect
> your workload to have the usual volume/surface-area scaling, and 
> thus doing a lot of work in a single node, and needing only moderate
> bursts of bandwidth for nontrivial problem sizes.

   Those who have experiences with fluent here (readers of the list) 
have mentioned it is latency sensitive.  I have some independent test 
data which shows this as well as my own benchmarks.  While I am familiar 
with where and how the numerical models in a (specific) CFD code would 
perform, fluent is a whole nother animal - as there are many things 
added to it to make it more stable, robust and generally 'user' 
friendly.  While, I'm sure they strive to make it run as fast as 
possible I'm sure this is not their number one goal.  Thus it has very 
different characteristics than a CFD code written for a narrower band of 
problems - as well as lacking the benefit of source code ;-)

Chris Sideroff

More information about the Beowulf mailing list