[mpich-discuss] SGE & Hydra Problem

[Pavan Balaji]

Great. I'll commit this in.

Based on Reuti's previous message, I remembered that qrsh can be configured to use ssh or rsh internally as well. With older openssh versions, there was a problem when the stdin socket is closed before calling ssh. Newer openssh versions seem to have fixed that. So, we had a work around in Hydra for ssh to handle this case, but only for ssh. I just extended that workaround to be used for sge and rsh as well.

So my guess is that the second cluster has a newer ssh version while the first cluster has an older ssh version and qrsh uses ssh internally.

 -- Pavan

--
Pavan Balaji
http://www.mcs.anl.gov/~balaji

----- "Ursula Winkler" <ursula.winkler at uni-graz.at> wrote:

> Pavan Balaji schrieb:
> > Can you try this patch:
> http://www.mcs.anl.gov/~balaji/tmp/sge.patch
> >   
> 
> Good news: I applied the patch and now it works - thanks you very
> much!
> 
> (I just still don't understand why it worked before in about 1:20
> trials 
> and why
> it has worked on the second cluster with the same configuration
> without 
> patch.)