[GE users] PE wildcards and reservations

John Saalwaechter johnsaalwaechter at yahoo.com
Fri Nov 11 19:09:34 GMT 2005


    [ The following text is in the "iso-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

We have created a sequence of related parallel
environments (PEs) for a certain application. This
allows a set of hosts to be chosen as a unit, based on
their network topology.  The PEs are something like:
foope_01, foope_02, ...

The app script then uses the PE wildcard mechanism to
get scheduled to a set of hosts (i.e. "qsub -pe foope*
4 ...").

I'm trying to use resource reservation so that the
hosts can be used by non-parallel BATCH jobs while not
processing parallel jobs.  Here's the problem (based
on $SGE_ROOT/default/common/schedule contents):

* BATCH jobs are streaming into all the hosts
* I submit multiple instances of the parallel
application with "-R y"
* SGE chooses to target the foope_02 (for example)
parallel environment for all instances of the parallel
job
* SGE does not spread the reservation target among all
the foope* available PEs, nor does it dynamically
retarget
* As a consequence, all instances of the parallel
application are serialized through foope_02, which is
a bad thing

Ideally SGE should spread the reservations across all
the PEs matched by the wildcard, and it should
probably have some heuristics about when to make
adjustments to what it is trying to reserve, instead
of making a static decision once.

This is N1GE 6.0u4, and I wasn't seeing anything
related in http://gridengine.sunsource.net/.

Is this a known issue or limitation?  Is it already an
RFE? Are there workarounds?

Thanks!

--
johnsaalwaechter at yahoo.com


		
__________________________________ 
Start your day with Yahoo! - Make it your home page! 
http://www.yahoo.com/r/hs

---------------------------------------------------------------------
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