[openib-general] mvapich-gen2 question

[Weikuan Yu]

On Sep 13, 2005, at 9:31 PM, Makia Minich wrote:

> I'm using a RHEL4 based system with the backport-2.6.9 svn drop 
> (svn3279).
> Building the mvapich-gen2 from subversion against this, everything 
> seems to
> be ok, and installing it goes well.  The problem is when I run a test I
> get the following error:
>
> :::<START>:::
> => mpicc -o osu-bw osu-bw.c
> => mpirun_rsh -rsh -hostfile ~/machines -np 2 ./osu-bw
> /benchmarks/osu/src
> /benchmarks/osu/src
> [1] Abort: Error creating CQ
>  at line 121 in file viainit.c

This means one of your node have some problems in allocating resources.
Please check the output of the following command

# ulimit

You may have a default, limited mlock limit, 32k for example. If so, 
please do these steps
a) un-comment the following line in /etc/limits.conf to remove memlock 
limit.
# *               soft    memlock         unlimited

b) And also put another line to the beginning of /etc/init.d/sshd to 
make it default for any new login.
ulimit -l unlimited

Please let us know if the memlock limit is the problem you are facing.

> mpirun: executable version 1 does not match our version 2.
>
> done.
> =>
> :::<END>:::
>
> I see in the code for mvapich (in ch-gen2) that there is a check 
> against the
> version, but I'm not quite sure where this version is defined in my 
> compiled
> code.  Perhaps there's something I'm just not seeing.

If the problem is not due to memlock limit, we will happy looking into 
this further. If possible, a temporary account that helps to reproduce 
the problem would speed up things significantly.

BTW, the version number here is defined to facilitate external process 
manager to check/match the protocol used at startup time. The actual 
code is defined in this file: mpid/ch_gen2/process/pmgr_client.h

  #define PMGR_VERSION 2

Thanks,
Weikuan

>
> Thanks....
>
> (((((((((((((((((((((((((((((((((())))))))))))))))))))))))))))))))))
>  Makia Minich                      Money is the Devil's toothpaste.
>  925.XXX.XXXX                              --The Flea (Mucha Lucha)
> (((((((((((((((((((((((((((((((((())))))))))))))))))))))))))))))))))
> _______________________________________________
> openib-general mailing list
> openib-general at openib.org
> http://openib.org/mailman/listinfo/openib-general
>
> To unsubscribe, please visit 
> http://openib.org/mailman/listinfo/openib-general
>