[GE users] RE : [GE users] RE : [GE users] SGE 6.2 : use of ssh for qlogin/qrsh

reuti reuti at staff.uni-marburg.de
Sun Nov 23 20:23:27 GMT 2008

Am 23.11.2008 um 20:36 schrieb igardais:

>> [...]But even then, you wouldn't have a Tight Integration.
> Hum ... well ...
> Can you shine a light on me and explain exactly what is considered  
> as tight integration and what is considered as loose integration ?
> I thought that starting remote mpd processes though qrsh was enough  
> to qualify for tight integration.

Yes - and no.

Yes, the forked process on a slave nodes has (as a result of the qrsh  
used by mpdboot) still the additonal group ID attached (needs the  
thight ssh integration with a special compilation flag, unless you  
use the default rsh or builtin startup).

And no: when the shepherd left the slave node, there is noone left  
taking care of these accounting data due to the additonal group id  
(cpu, mem and io). A qdel will still kill the keft behind escaped  
process (although I don't know why - the group id could have been  
reused in the meantime - maybe it's a feature of MPICH(2)). The ru_*  
accounting data is missing by principle, if only the processes select  
by the additional group are used.

Furthermore: what happens with two jobs of the same user on a node?  
You end up with only one daemon running and an "mpdallexit" will  
bring both jobs down.

Tight Integration means: qdel works and affects only the processes  
belonging to a job. Accounting data is correct. After the job all  
daemons are gone again.

-- Reuti

> Ionel
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do? 
> dsForumId=38&dsMessageId=89612
> To unsubscribe from this discussion, e-mail: [users- 
> unsubscribe at gridengine.sunsource.net].<winmail.dat>


To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].

More information about the gridengine-users mailing list