On Feb 27, 2008, at 2:19 PM, Ethan Mallove wrote:
We were seeing NetPIPE runs timing around 45sec the other day, so I set the timeout value to -1 (which is infinity), and last night a v1.2 NetPIPE run took 3 hrs!
Yoinks!
It looks like the test got stuck: 0: burl-ct-v440-2 1: burl-ct-v440-2 2: burl-ct-v440-2 3: burl-ct-v440-2 Now starting the main loop 0: 1 bytes 5 times --> 0.49 Mbps in 15.59 usec
You might want to run this by hand and see what's going on. Don did make some udapl btl changes recently.
The other bizarre thing is that NetPIPE now passes with the infinite timeout, but seems to take the *same* duration as the timed out runs: http://www.open-mpi.org/mtt/index.php?do_redir=539 (I was expecting NetPIPE to take *longer* than 45sec.) Any ideas why this would be?
Not offhand, no. I'd run them by hand and see what's going on... -- Jeff Squyres Cisco Systems
