Missing O3AS-tasks urgently
4 Oct 2023 8:48:31 UTC
Topic 230161
(moderation:
Hallo!
Why don't I get any more tasks for 03AS?
Work generator is running already and in stock are more than 46K tasks and I have no tasks faulty crunched.
Kind regards and happy crunching
Martin
From you scheduler reply, you
)
From you scheduler reply, you aren't asking for O3AS work. Check your preferences and venues.
astro-marwil wrote:Why don't
)
Just over a week ago, you downloaded close to 600 O3 tasks. Less than an hour later you seem to have aborted all that remained. If you wanted those tasks, why did you abort them all?
They may not be compute errors but the scheduler will use aborted tasks to decrease your allowance in just the same way as for compute errors. Maybe you ended up in the 'sin bin' for a while :-).
You now seem to have a whole bunch of BRP7 tasks. I hope you're not planning to abort them as well :-).
Cheers,
Gary.
Hall Gary!Thank you for
)
Hall Gary!
Thank you for your answer.
As you said, I got a big batch of tasks when I started O3AS. But this batch was far too big to be crunched within the given time limits. So I send them back, when a became aware of this. And I was right. The remaining tasks I finished just within the time limits given. That I got too many tasks is a well known problem of E@H, as they do calculate in an unrealistic way the foreseen crunching time. For O3AS they promised a much too little crunching time. For BR7 it was just reverse. They promised about 2h instead of really 12.25 min.
This problem is well known, as I know from others in this forum.
So I think, it's not my fault what happened! And there is some room for improvement of E@H program in some way.
Kind regards and happy crunching
Martin
astro-marwil wrote: As you
)
Astro-Marwil,
The reason that "others in this forum" are commenting or even complaining about how many tasks they receive vs the ones that they can actually do is (likely) because they don't have a config file in their project folder or haven't got it configured properly yet. I'm referring to Einstein, of course, but it just may apply to many, if not all projects.
Have you an "app_config.xml" file?
https://boinc.berkeley.edu/wiki/Client_configuration#Application_configuration
I refer to these two comments within the app_config.xml file that you may or may not have.
max_concurrent - The maximum number of tasks of this application to run at a given time.
fraction_done_exact - if set, base estimates of remaining time solely on the fraction done reported by the app.
I'm just sayin'... Lighten up!
Proud member of the Old Farts Association
You are asking for too much
)
You are asking for too much work. Reduce your cache size. There is no penalty for only keeping a few tasks onboard.
When you finish your existing work, the client will ask for and download a few more for you to crunch.
If you got 600 O3AS tasks the first time, reduce your cache size by 100X so you only get 6 tasks.
astro-marwil wrote:... That I
)
If the problem is so "well known", why did you not take some precautions? What is more "well known" is that the previous FGRPB1 search was indeed much faster than the initial estimate, particularly for people with modern GPUs. You were previously running that search so you would have had a very low duration correction factor (DCF). The mistake you made was to not take account of that before starting a different search. It's pretty easy to stop BOINC and edit the state file with a plain text editor to reset the DCF to 1.000000 (the default value) before starting a new search
There's no need to even do that if you think about it. Before starting any new search, you could just set your work cache size to a very low value. For example you could set the two values to 0.01days/0 days and no matter what the initial estimate was set to you would get just 1 task for each device that was going to crunch it. When that task arrived, you would then see the initial estimate and when each task finished, you would then know the true crunch time.
That first task will reset the DCF to a more appropriate value. If you are setting your work cache size locally in BOINC Manager, and having seen the corrected (and much longer) estimate, you can immediately change the work cache size to the value you want.
They don't 'promise' anything. They just make an educated guess, which could be quite wrong. There are likely to be substantial variations between all the different generations and brands of hardware and software out there. They don't have the staff or the equipment to pre-test everything in advance so they are forced to to choose what they think will be 'average'. Each volunteer should test that for themselves. Always ask for small amounts of work when you are starting a new (or unknown) search.
Cheers,
Gary.
GWGeorge007
)
Thanks, I previously missed this option.