On Wed, 4 Sep 2002, Ricardo Fonseca wrote:
> I've been trying LAM (6.6b2cvs) on MacOS X 10.2 (Jaguar release, build
> 6C115) using the latest developer tools (aug. 2002 update). All rpi's
> (except Myrinet GM) work ok on a single node, since jaguar now includes the
> proper semaphores; however, when I try it on two different computers I get
> very poor bandwidth:
We're actually just now getting our Apple machines up and running under
Jaguar. We haven't even got to real LAM testing yet :(.
> (yes, the tcp rpi is actually MUCH slower than the lamd). These results are
> average results for a 1Mb message, and they varied a lot for the 10 messages
> tested. This is running on a 100 Mb/s network. Tests with lam 6.5.6 on the
> same network gave about 11.7 MB/s with very low variance (with other
> machines).
Very strange. I can't thing of anything that would cause that great of a
variance in the CVS version but not cause the problem under 6.5.6. Like I
said, I haven't gotten to running tests of LAM on Jaguar yet, other than
to notice that SysV semaphores seemed to work properly.
At this point, I wouldn't assume the problem is in your machines. I
wouldn't rule that out, either :). I'll try to run some tests between my
iBook and my G4 tower and see if I can come up with anything. Obviously,
I'm not going to get great numbers out of that combo, but I should be able
to see the problem you are describing...
Brian
--
Brian Barrett
LAM/MPI developer and all around nice guy
Have a LAM/MPI day: http://www.lam-mpi.org/
_______________________________________________
This list is archived at http://www.lam-mpi.org/MailArchives/lam/
|
