Duplicate Assignments?

[nofaith628]

Today I got an unexpected error from MISFIT which wrote 'result was not needed'. See capture1.jpg.

These assignments were automatically assigned via MISFIT from GPU to 72.
And indeed, it looks like 'monst' has already taken the assignment and completed it 4 days before my computer got to it.

See:
89429117
89430703

Which was a rather weird event to see. So I went over to the GPU to 72 website to check if the exponents were assigned to me, and they were. See capture2.jpg

Are there any complications with the GPU to 72 website and mersenne.org lately? Thinking the system might have assigned two users to one exponent.

Edit: GPU to 72 assigned the exponents to my computer after it has already been factored, could the system admin take a look into this issue?

[0PolarBearsHere]

I'd guess that it's part of the recycling script http://www.mersenneforum.org/showpos...7&postcount=79

[Mark Rose]

Besides the assignment recycling issue, where old assignments eventually get turned in (which I've done accidentally before, my bad), there are some users who look for "oddities" in the mersenne.ca status charts or elsewhere who will TF without getting an assignment from GPU72 or GIMPS. The only good way to counteract that, I've found, is to keep a short queue of TF work. If working on a large batch, I'll try to submit all at once to not leave any "oddities" to be poached.

[kladner]

Originally Posted by Mark Rose

Besides the assignment recycling issue, where old assignments eventually get turned in (which I've done accidentally before, my bad), there are some users who look for "oddities" in the mersenne.ca status charts or elsewhere who will TF without getting an assignment from GPU72 or GIMPS. The only good way to counteract that, I've found, is to keep a short queue of TF work. If working on a large batch, I'll try to submit all at once to not leave any "oddities" to be poached.

To elaborate, monst has something of a record for this sort of thing. I tried to find the thread in which someone in addition to me reported work taken by monst, but it is a ways back and I probably haven't looked in the right places.

In any case, nofaith628, you are not alone.

[chalsall]

Originally Posted by nofaith628

Edit: GPU to 72 assigned the exponents to my computer after it has already been factored, could the system admin take a look into this issue?

Ah, shit... Sorry nofaith628.

GPU72 officially reserves assignments from Primenet which are not already reserved, and only then "lends" them out workers to work.

It appears that monst is also working this range, but not reserving the candidates. This is a bit strange since monst is also a GPU72 participant.

The good news is monst seems to be only taking candidates to 72 bits in the 89M range, while GPU72 requires commitment to 73 bits. Thus, for every candidate he did and you also did which didn't find a factor, you only wasted 1/3rd of your effort, and you still all get all the credit.

But your last point is valid... The system should not have issued an assignment to you after it was already factored. I will have to do a deep dive into the code and the database to figure out what exactly happened.

Thanks for the extra work monst.

Oh well, as the saying goes, no good deed goes unpunished....

[chalsall]

Originally Posted by 0PolarBearsHere

I'd guess that it's part of the recycling script http://www.mersenneforum.org/showpos...7&postcount=79

Nope. GPU72 no longer recycles TF nor P-1 assignments and reassigns them to other GPU72 workers.

Instead I spend the money to finish them. And if the assignee wants to yell at anyone, they can yell at me for "poaching" an assignment which is overdue.

This situation is different -- this work was done without reservation.

[nofaith628]

Originally Posted by chalsall

But your last point is valid... The system should not have issued an assignment to you after it was already factored. I will have to do a deep dive into the code and the database to figure out what exactly happened.

Thanks for the prompt reply chalsall, glad to know the situation is under control.

After second thought, how can someone manually submit TF results that have already been reserved by 'GPU Factoring'? Can users still manually input assignments into their worktodo.txt and submit the results disregarding that the fact the exponent has already been assigned?

[Mark Rose]

Originally Posted by nofaith628

After second thought, how can someone manually submit TF results that have already been reserved by 'GPU Factoring'?

By using the manual submission form or a script/program that does the same.

Can users still manually input assignments into their worktodo.txt and submit the results disregarding that the fact the exponent has already been assigned?

Yes. If reserving from PrimeNet they won't get an assignment ID, but they can still do the work and the server will accept the result (except for duplicate TF work where the first submission wins).

[monst]

Sorry about this everyone. I had grabbed all the 89M candidates at 71 bits a while ago (2 or 3 months) before GPU72 reserved them. I was taking them to 72 bits. I had finished about 40% of them when I noticed that GPU72 reserved them. When I noticed this I stopped immediately, but my GPUs were crunching on them for a day or two before I noticed. I'm not sure how many were factored and how many were taken to 72 bits after GPU72 reserved them. My GPUs were sending results back to the database every 15 minutes so I did not have a large backlog of unreported work. Chris, do you know how many 89M candidates at 71 bits GPU72 reserved when it made the reservation?

[chalsall]

Originally Posted by monst

Chris, do you know how many 89M candidates at 71 bits GPU72 reserved when it made the reservation?

Hey Monst. Thank you very much for your response. Mistakes happen -- this was an honest mistake by both of us.

I don't know exactly how many candidates were reserved in 89M at 71 bits by GPU72. I could probably figure that out (I think the information was captured) but I don't have an immediate SQL query ready to run.

In the future, if you're going to work ahead of the "wave" without reserving from Primenet, perhaps we could coordinate.

Also, separately, I have a bug in my code. The system should have detected factors found before assigning work. I haven't yet figured out why that subsystem didn't work properly.

[nofaith628]

Pinging chalsall.

Another batch of duplicate results popped up. See the attached pictures, I have pointed out the reserved assignments on GPU72 website in the red blocks.