Duplicated credit

[ckdo]

Originally Posted by Unregistered

What I don't like about this is knowing that anyone can make other people waste time in an assignment that will be rejected.

What I was trying to say is that, at the far high end of PrimeNet's exponent scale, it becomes infeasible for the people doing 99.999% of the work there to respect single assignments by others. They have a hard enough time not stepping on each others' toes. Credit goes to whoever finishes an exponent first, not whoever reserved it.

Yes, this could be changed. If you reserved an assignment you could get the credit even if someone else completed it. That would then lead to people reserving exponents and never processing them, waiting for free credit when someone else did, probably.

Or if you have reserved an assignment and someone else finishes it, they couldn't get any credit at all (ignoring the result alltogether, eventually), and you wouldn't get the credit until you actually finish the assignment yourself. Now since whoever reported the result supposedly did the work, too, is it fair not to credit them?

Originally Posted by Unregistered

Also, don't forget the double-crediting problem.

As far as ECM is concerned, the problem of potential double credit can not easibly be avoided. A single exponent may require hundreds of thousands of curves at various bounds to be run on it in order to come up with a factor.

People will report that they have done this and that many curves at some bounds or other, and if they can back up that claim with the correct checksum for the task, they will get the credit. If they later run the same amount of curves with the same bounds on the same exponent, the checksum will be identical. There's no way of telling whether someone actually did the work twice or is just reporting the result twice.

Consider the analogy with bank notes: If you hand me 20 one dollar bills, I will usually trust you that no two of them have the same serial number. Now even if I kept track of all bills I ever had, I will probaly not call the police if you come back a year later to hand me another 20 one dollar bills and I find that I had already received one of them the year before. I would assume that by whatever means you got the exact same bill back, and accept it.

Likewise, the "serial number" for an ECM curve is the combination of exponent, bounds, and s-value. Even if you reported your individual s-values along with the rest and the server kept track of all of them to identify duplicate curves, what would you expect to happen? You'd report 20 curves and be credited 19. You may still have come up with the same s-value by coincidence, in which case you'd be missing credit for work you truly did. The only ways of making sure that no s-value is reused are either transferring the entire list of used s-values to the client or the client asking "is this s-value okay?" before it actually does the work, neither of which is even remotely feasible in terms of storage space, bandwidth, and CPU time, as well as breaks compatibility with all existing software.

Originally Posted by Unregistered

Is it that much to want these little things?

These "little things" are actually quite insurmountable obstacles. Either you double credit work, which would increase the potential of cheating, or you don't credit work which actually has been done. Neither is a good choice, and depending on the type of work, different (non-) solutions to the problem have been found.

As long as you stay on the beaten track, you're mostly safe. Deep in the woods, however, you may encounter bears, and the rangers may not be there to shoot them. As far as PrimeNet is concerned, both ECM and (TF-) LMH should be considered the woods.

Occassionaly, a bear may even shit, err, cheat in the woods. That issue is best ignored. There's a number of people around who will double check unrealistic results, identifying both software bugs and cheaters in the process.

Thanks for the detailed explanation, ckdo. I should have been more clear when I said "little things". It's probably quite a challenge avoiding the manual duplication before it happens, but perhaps it would be more viable allowing the user say to the server "hey, those results are duplicated, please remove them" once they are submitted. Another way could be telling George which results are duplicated, but I suppose that would be too time consuming.

[bloodIce]

I was really astonished to see my nickname engaged in this discussion. Firstly, I am sorry to Unregistered for what I seem to have caused. I have an explanation. I remember that during a day in August I was checking exponents in some range, I guess the mentioned range unaware that someone else is working in the same range. I think I noticed that I overlapped with someone so I aborted the whole schedule for my program. Than I checked what was the current report for current range. I found 3 or 4 exponents which were not factored to the border (64 I assume). I was surprised that whoever was working there missed those. So I run those (which took me around 10 min) and I was really surprised that the first returned a factor, second returned a factor... and up till the last. The factors were submitted automatically after they were found. Well, I would like to say sorry for that. I do not know what to do now with the issue. I could agree that those 3 or 4 exponents belong to Unregistered and if it is possible they could be stripped from my record. P.S. The exponents I was talking about were TF, not ECM efforts, so the time spend on those was less than 5 min probably. 848684891 and 848672197 are still not factorized and yes I was TF-ing them to 64. However, I could not see which are the exponents in question 848171213, 848171747, 848195707 and 848198129 may be?