Re: Re: MPICH-GM problem: cannot run multiple executables in one job

Hi Alan,

On Thu, 2003-03-20 at 07:44, Alan Scheinine wrote: 
> I do not know if "--gm-wait 1" was really the solution, there may have
> been other factors such as the load of the nodes during the various
> times that I ran the tests, but it may be worth a try.

The only reason I see why adding a wait time of 1 second between
fork/spawn steps would make a difference is that your rshd or sshd
service have a DoS attack protection and it limits the number of
incoming connection per minute. If you connect to a machine with this
feature enabled, the first incoming rsh/ssh connection will be accepted
and they will hang once the threshold is reached. You can increase this
limit by adding a (large) number to the "nowait" keyword of the shell
entry in INETD conf file like "nowait.200"

>    The option "--gm-wait n" is an option for mpirun when ch_gm is
> being used, it is not necessary to change the mpirun.ch_gm.  I changed
> the file mpirun.ch_gm because I was running the built-in tests of mpich.

You should be able to pass args to mpirun via the "-margs" flags of the MPICH 
test suite run scripts.

Patrick
-- 

Patrick Geoffray, PhD
Myricom, Inc.
http://www.myri.com