These 6 computers have been on Primegrid for a month now, so my computers know how long tasks take, and the times are shown correctly. But even doing short tasks (say genefer 15 on a GPU) which take 3 minutes each, with a 3 hour buffer I'm given 8 hours of tasks. Why? This really screws things up with a longer buffer, because too much is downloaded and other projects fail to meet their shorter deadlines. For example, I set the buffer to 3 days, prime gives me 8 days, there's too much work, panic mode is engaged, and we all know panic mode in boinc doesn't work properly, it leaves things to the last minute and when it has 24 core prime and many 1 core WCG tasks to do, it can't work out how to fit everything in sensibly. WCG tasks are getting returned late. Please fix the server to give out the amount of work I request, not 3 times as much!

BOINC can be quite unpredictable.

Have you considered having zero buffer, for projects (including Genefer) where there is a "race" and you will be declared finder of a prime only if you win that race (return your task 1st)? This does require that your computers have permanent connection to the internet.

/JeppeSN

No one wants to be DCer, right?

I think I read somewhere that the problem is within the Boinc software, not the server.
So it might be useful to try other versions of Boinc.
____________
Greetings, Jens

92914140^65536+1

I think I read somewhere that the problem is within the Boinc software, not the server.
So it might be useful to try other versions of Boinc.

Change the Job Control setting from unlimited to a reasonable maximum number of tasks.
____________

No one wants to be DCer, right?

Weeeeellllllll maybe...in todays World of someone always being able to get faster and better hardware being the DC'er is much better than being nothing at all. Nothing at all being those that aborted the task or for some reason the task errored out and had to be resent to someone else. NO I'm not denigrating those that the task times out on or choose to abort the task for whatever reason or those that for some reason the tasks fails to complete correctly...they are a fact of crunching.

I agree with Jeppe's solution:
for non-LLR2 projects (WW, AP, all GFNs, PPSE, SGS), it is highly recommended that you DO NOT have a bunker of more than 0.

These tasks are sent out in magnitudes of two, and in the case of a prime, the user who crunched the first returned task will be primary finder and the second user double checker. If two users have the same computing force, and one user has a bunker, then the user who doesn't will process that task immediately while the one with a bunker will most probably process some other tasks before getting to this prime one. And therefore resulting in being double checker.
No one wants to be DCer, right?