[Beowulf] Network benchmark - MPI

[Tony Travis]

Jonathan Michael Nowacki wrote:

> Tony,
> 
> I'm actually out at the MBL for a summer internship and I just got word
> that they have experienced the 4 node thing as well.  Thought you should
> know.

Hello, Jon.

Thanks for the info: I've not finished my testing yet but here are some 
results for a few different numbers of processors(p) and chains(c):

   1p4c:      Chain completed in 480051 seconds
   2p4c:      Chain completed in 266443 seconds
   4p4c:      Chain completed in 189369 seconds
16p16c:      Chain completed in 221835 seconds

I'm still running the 8p8c test which I've had to restart several times 
because of other work on the cluster. However, I expect it will take 
longer than 4p4c...

> More detailed info on our clusters @
> http://jbpc.mbl.edu/computing/hardware.html

Quite interesting :-)

> The nodes only have 512MB memory and are going through about 2 gigs of
> swap, but we are upgrading them.  My boss says MrBayes is probably limited
> by the default 4 Markov chains, and thats why their is no significant
> gains after 4 nodes.  I wonder if quad processor nodes would speed it up,
> or if the actually speed of each processor is the only thing that would
> matter.

There seem to be other algorithms that work 'best' with four processors, 
  but I wonder if it's just a limitation of our 100Base-T network fabric?

	Tony.
-- 
Dr. A.J.Travis,                     |  mailto:ajt at rri.sari.ac.uk
Rowett Research Institute,          |    http://www.rri.sari.ac.uk/~ajt
Greenburn Road, Bucksburn,          |   phone:+44 (0)1224 712751
Aberdeen AB21 9SB, Scotland, UK.    |     fax:+44 (0)1224 716687