Also immer rechtzeitig abschicken und nicht zu lange bunkern ...:
----------------------------------------------------------------------
Hi all,
Some members have recently noticed that WUs appear to be aborting before the WU has completed or have seen a WU abort at around 100 hours of CPU time.
I can now confirm this is intentional behaviour and the UD agent does not have a problem. The reason for this is explained below...
With Agent version 2.1, UD added a feature to enable time limits to be set on WUs. The main reason for this was to try and limit the impact that Long Running Molecules (LRM) were having on the project. Initially the time limits were not used but this has now changed and current WUs now have time limits defined.
UD have now added 2 time limits to a WU when it is received by the UD agent: CPU time and Wallclock time...
CPU Time: This imposes a maximum CPU time on the WU. If a WU takes longer than this time to process it will be aborted and returned as an incomplete result.
Wallclock Time: This imposes a maximum time that the WU is valid for. This means the WU MUST be processed within this time. If it is not processed or this time expires while a WU is being processed, the WU will be aborted and an incomplete result will be returned.
Different proteins may have different timeouts associated with them so it is not possible to determin when running what CPU or wallclock limits exist for a WU. As a guide the CPU time limit will be approx 5 to 7 days (100-150 hours) and the wallclock limit around 12 to 13 days.
In addition to the above version 2.1 also added a feature to allow credit to be awarded to aborted or incomplete WUs. In the above cases (and when a WU is killed) you will now get credit for the CPU time spent on the WU and points related to the CPU time. You will not however get a credit for a result returned as it was incomplete
UD Monitor Users
----------------
Because WUs now have a predefined life (wallclock limit above), if any WUs in cache slots are older than 12-15 days they will be regarded as expired and will not be processed. This will result in the WU aborting as soon as UD Mon switches to that cache slot.
To reduce the likelyhood of this happening you will need to reduce the number of cache slots to a level where no WUs are more than 12 days old.
Note1: Incomplete results are of no value to UD, Oxford or this project so they are discarded.
Note2: When the UD agent aborts a WU it will trigger the download of the THINK program again (approx 600k) plus a new WU.
Note3: The above does not solve the save issues but will limit the time processing a WU.
Cheers
Orbi
----------------------------------------------------------------------
... unser armer Teamgruender GTFreak kann ein Lied davon singen,
denn die folgende Neuerung vom 15.11.2001 konnte er nicht nutzen:
----------------------------------------------------------------------
We have doubled the current timeout periods to allow for the longer set of WU's that are in the current job. The next set of workunits you receive will have a timeout of 200 hours for CPU time. Thanks for the info.
Moose
----------------------------------------------------------------------
Wer sich fuer weitere Infos interessiert der klicke mich.
Happy crunching and welcome back the 6COX-LRMs 
[Diese Nachricht wurde nachträglich bearbeitet.]
