[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]


On Wed, Jan 28, 2009 at 09:23:26AM -0600, Todd Tannenbaum wrote:
> Steffen Grunewald wrote:
> > For a homogeneous pool, and "simple" job clusters (identical specs for all
> > jobs) NEGOTIATE_ALL_JOBS_IN_CLUSTER is suggested to be set to False.
> > On the other hand, there may be situations where the first job of a single
> > cluster continues to fail (for whatever reason: memory overcommit comes to
> > mind) thus blocking all others.
> Hi Steffen  -
> What version of Condor are you working with?

7.2.0, previously 7.0.5

> Starting back w/ Condor v7.0.x and above, the default built-in auto 
> clustering mechanism in Condor should prevent the situations you 
> describe above --- and do so in a much more efficient/scalable manner 
> than setting NEGOTIATE_ALL_JOBS_IN_CLUSTER to TRUE (which is the kiss of 
> performance death if you have thousands of jobs).

This means I shouldn't touch the default setting then?

> > Is it possible to - e.g. once per given time period (4 hours?) - "flush"
> > the queue by temporarily setting the macro to True?
> >
> Maybe something else is going on?  With Condor v7.0.x and above with the 
> default auto-clustering, I assert you should never have to resort to 
> NEGOTIATE_ALL_JOBS_IN_CLUSTER = True.    Are you over-riding 
> autoclustering in your config file by expliciting setting 
> SIGNIFICANT_ATTRIBUTES or some such on your condor_config on your submit 
> hosts?

No, and I never considered to.


Steffen Grunewald * MPI Grav.Phys.(AEI) * Am M�erg 1, D-14476 Potsdam
Cluster Admin * http://pandora.aei.mpg.de/merlin/ * http://www.aei.mpg.de/
* e-mail: steffen.grunewald(*)aei.mpg.de * +49-331-567-{fon:7233,fax:7298}
No Word/PPT mails - http://www.gnu.org/philosophy/no-word-attachments.html