Message boards : Number crunching : variable run time
Joined: 30 May 06
I saw a post about the twisty units getting better results if you ran them at 6 hours vs 4 hours.
I ran the batch I had at 6 hours but then that set my other work behind schedule and even with setting things back to 4 hours I ran behind schedule and then I aborted the past the deadline work because I figured I would not get any credit for it.
Now, if my normal base run time is 4 hours and then I set it for 6 hours to get good results on specific work and then set the run time to 2 hours to catch back up on deadline and then set the run time back to 4 hours how is this going to affect the credit and the science result?
Joined: 17 Sep 05
Each time you change the run time setting it will confuse the scheduler on your hosts. I would advise setting it to the longer time and leaving it there.
BOINCing since 2002/12/8
Joined: 22 Aug 06
The science is uneffected either way. The recent questions have been more related to tasks that require a higher then normal minimum to complete a single model. Some taking more then 6 hours to complete a single model, which is the minimum amount of useful work. And so if you set your preference to 6 hours, it causes these tasks to complete in roughly the time you expect. The same task, run by a machine with a 2hour runtime preference is still going to take the 6 hours to complete, and hence not be completed in the time expected.
Yes, changes in target runtime effect the tasks you already have downloaded, and can confuse the scheduler (and possibly aid in meeting deadlines).
In general, when adjusting runtime preference, make changes gradually. And make changes when you do not have a large backlog of work downloaded. BOINC just knows how long the last task took. It really knows nothing about how long any uncompleted tasks will take.
Rosetta Moderator: Mod.Sense
Message boards :
Number crunching :
variable run time
©2023 University of Washington