[Swift-user] Teraport question

Mihael Hategan hategan at mcs.anl.gov
Thu Mar 19 19:06:37 CDT 2009


On Thu, 2009-03-19 at 18:29 -0500, Michael Wilde wrote:
> 
> On 3/19/09 6:22 PM, Mihael Hategan wrote:
> 
> >> "If no times are specified, but a queue is specified for the site, then 
> >> the wall time allowed the worker job is determined by the local 
> >> scheduler, based on the requested queue."
> > 
> > I have no idea where that came from.

I thought the paragraph was in the user guide. My answer was to that.

If it's your paragraph, then that's not happening because I don't know
how that could be done (i.e. figuring out the local queues and their
walltime limits).

> 
> It came from this reasoning:
> 
> Ben suggested to Yue a sites.xml entry with a queue spec, and I assumed 
> the intent was to use that without a wall time.  I then further assumed 
> that if this worked (and Yue indicated that it did: his jobs wound uop 
> in the fast queue), then the jobs must be getting some default time spec.
> 
> Ive always in the past found it a nuisance to have to get the 
> maxwalltime spec compatible with the queue name, as you have to hunt 
> town this data for every site.

Yes. This is an issue that The Grid introduces. If you have one cluster,
you read the documentation for it. If you have a bunch of clusters
coming and going, you hit keyboard with hammer.

> 
> If you knew the queue names, and could just specify a queue name without 
> a wall time, that seemed handy. (Although I suppose can lead to the 
> pitfall of unexpectedly exceeding wall time limits).

This let's maybe or like perhaps put job in possibly queue seems a
little fishy, given that coasters rely on knapsacking job walltimes into
worker walltimes to get things done.




More information about the Swift-user mailing list