You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As discussed in Freiburg, could we have jobs submitted to the CE with say
+site='LRZ-LMU'
only go to drones running on this site?
This would allow us to have multiple Panda queues, in the same overlayBS, each using local storage for stage-in/out.
The text was updated successfully, but these errors were encountered:
This would let us use the data-colocation capabilities of Panda. Maybe slightly more generic would be to accept some information about the input data location and size, then use this in the scheduling decision.
For example, we could have a topology saying DESY-HH is close to GoeGrid_NHR and the latter operates an XCache with some chance of the data being cached. Then build a Rank to prefer HH but it can run at NHR.
With scheduling decision, do you mean on the ATLAS level that they send pilots to a specific subsite, or do you mean on the Overlay Batch System (OBS) level?
Send jobs to a specific subsite needs to be handled by the OBS. TARDIS will just react to the utilization of the subsites which is a result of the decision made by OBS.
As discussed in Freiburg, could we have jobs submitted to the CE with say
+site='LRZ-LMU'
only go to drones running on this site?
This would allow us to have multiple Panda queues, in the same overlayBS, each using local storage for stage-in/out.
The text was updated successfully, but these errors were encountered: