[Beowulf] Problems scaling performance to more than one node, GbE

[Nicholas M Glykos]

Hi Tiago,


<snip>
> I tried with Gromacs ...
</snip>

Concerning your MD tests, would it be worthwhile to check also NAMD's 
parallel efficiency ?  Based on past experience, I would suggest that you 
try both the UDP- & TCP-based versions (don't forget to use the +giga and 
possibly the +atm flags). Also, keep in mind that parallel efficiency 
varies greatly with problem size (for example, you should expect very 
different results from, say, a 1,500 atom peptide system, as compared with 
a 100,000 atoms large protein system).


> From what I could also digg around, it seems that some switches have too
> much latency and hamper any kind of proper performance from GbE. Were this
> my case, any benchmarks I could use to test that theory out?

We have been using NetPipe (http://www.scl.ameslab.gov/netpipe/) to 
measure latency, but the experts on this list will probably tell you more 
about the subjest that you'd ever wished to know ;-)


Nicholas

-- 


             Dr Nicholas M. Glykos, Department of Molecular 
         Biology and Genetics, Democritus University of Thrace,
   University Campus, 68100 Alexandroupolis, Greece, Fax +302551030620
      Tel ++302551030620 (77620),  http://www.mbg.duth.gr/~glykos/