[GE users] setting up mpich2 pe + qrsh

jeroen.m.kleijer at philips.com jeroen.m.kleijer at philips.com
Fri Feb 11 16:19:45 GMT 2005


Reuti,

I thought this would be the solution but was kind of wondering whether 
there were any others.
I'll be using this solution.

Again, thanks for all the help.

Met vriendelijke groeten / Kind regards

Jeroen Kleijer
Unix Systeembeheer
Philips Applied Technologies









Reuti <reuti at staff.uni-marburg.de>
2005-02-11 04:27 PM
Please respond to users
 
        To:     users at gridengine.sunsource.net
        cc:     (bcc: Jeroen M. Kleijer/EHV/CFT/PHILIPS)
        Subject:        Re: [GE users] setting up mpich2 pe + qrsh
        Classification: 




Jeroen,

for the admin login you could use ssh from the master to the nodes, and 
add a 
line like

AllowUsers reuti jeroen

to the /etc/ssh/sshd_config on the nodes.

rsh between the nodes for the users, ssh for the admins from the master to 
the 
slaves.

BTW: since qrsh is using a private daemon, rsh could be disabled 
completely. 
Problem may be the proper shutdown of the smpd on the slave nodes in case 
of an 
qdel, as you can only use the last way I mentioned as an option in 
stop_proc_args to shutdown smpd instead of the rsh command.

Cheers - Reuti


Quoting jeroen.m.kleijer at philips.com:

> Hi Andreas,
> 
> I've figured out what the problem was (forgot the check "control slaves 
= 
> true" in the PE environment) so that was sorted out.
> I am kind of wondering how a setup would look like where users are not 
> allowed to directly login to an execution host but are allowed to do a 
> qrsh -inherit -V.
> 
> My current setup was:
> /etc/nsswitch.conf
> passwd: compat
> /etc/passwd (distributed via NIS)
> (last entries)
> + at sysman:x:::::
> +:x:::::/bin/false
> 
> This would allow only system administrators to directly login to the 
nodes 
> and anyone that's not in the sysman netgroup gets denied access.
> This is perfect for regular jobs but MPI jobs which require being able 
to 
> qrsh are denied access as well so I've had to set /etc/nsswitch.conf to:
> passwd: files nis
> 
> In this setup rsh is allowed (but using hosts.equiv, only from other 
> execution hosts)
> 
> Do you have any suggestions how to tackle this kind of situation? (I 
know 
> this is more system administration related than SGE but still)
> I want:
> allow direct acces to execution hosts for system adminstrators 
(preferably 
> via netgroups)
> deny direct access to execution hosts for regular users
> allow qrsh from one execution hosts to another via -inherit
> 
> Any suggestions would be greatly appreciated.
> 
> 
> Met vriendelijke groeten / Kind regards
> 
> Jeroen Kleijer
> Unix Systeembeheer
> Philips Applied Technologies
> 
> 
> 
> 
> 
> 
> 
> 
> 
> Andreas Haas <Andreas.Haas at Sun.COM>
> 2005-02-11 02:19 PM
> Please respond to users
> 
>         To:     users at gridengine.sunsource.net
>         cc:     (bcc: Jeroen M. Kleijer/EHV/CFT/PHILIPS)
>         Subject:        Re: [GE users] setting up mpich2 pe + qrsh
>         Classification: 
> 
> 
> 
> 
> On Fri, 11 Feb 2005 jeroen.m.kleijer at philips.com wrote:
> 
> > Hi Reuti,
> >
> > Sorry to keep bothering you with this but you (along with Ron Chen,
> > Andreas Haas and a host of others who I can't possibly name all) seem 
to
> > be the most active on this mailing list. (for which I am really 
> grateful)
> > I checked and the $TMPDIR (which is /volumes/scratch/<jobid>.batch.q) 
is
> > created on the starting host of the job (usually the nlcftcs14). This
> > directory doesn't get created on the other nodes (nlcftcs12 or 13),
> > neither by SGE itself nor the startmpi.sh script.
> > I'll comment out the mkdir entry.
> 
> Hi Jeroen,
> 
> the $TMPDIR must be created at each execution node. Code for this can
> be found in daemons/execd/exec_job.c near the comment
> 
>       /* make tmpdir only when this is the first task that gets started
>          in this queue. QU_job_slots_used holds actual number of used
>          slots for this job in the queue */
> 
> those $TMPDIR's however(!) get not created before the first task is
> started at an execution node. Possibly your setup depends on $TMPDIR
> be available before the first task is started?
> 
> Regards,
> Andreas
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
> 
> 
> 



---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net





More information about the gridengine-users mailing list