This is weird... Isn't it?

[guido72]

Is this the same result or not?...

[TheMawn]

The ECM method isn't deterministic. You don't get a yes-or-no answer from running a single curve. Part of what defines each curve is randomized so no two are likely to be the same. You get increased probabilities of finding a factor (if one is there) by running more. The user probably requested / submitted their curves in batches of 100 which is a bit odd since I think you have seriously diminishing returns at that curve count.

[guido72]

Originally Posted by TheMawn

The ECM method isn't deterministic.

I know that. My question was about how Primenet may distinguish from two (or more) different results if all the parameters (number of curves, boundaries, ...) it gets are the same... Or it knows that each curve is unic so that each result is unic as well...

[VBCurtis]

Originally Posted by TheMawn

The ECM method isn't deterministic. You don't get a yes-or-no answer from running a single curve. Part of what defines each curve is randomized so no two are likely to be the same. You get increased probabilities of finding a factor (if one is there) by running more. The user probably requested / submitted their curves in batches of 100 which is a bit odd since I think you have seriously diminishing returns at that curve count.

100 curves is nowhere near diminishing returns for B1 values above 50,000. at B1=110M, something in excess of 10,000 curves are called for before changing B1.

Primenet has no way to know the user isn't submitting fraudulent curve counts. Luckily, the nature of ECM is that a future user running curves of the same or larger size will usually find any factor the fraudulent user "would have" found.

[Madpoo]

Originally Posted by guido72

I know that. My question was about how Primenet may distinguish from two (or more) different results if all the parameters (number of curves, boundaries, ...) it gets are the same... Or it knows that each curve is unic so that each result is unic as well...

Hard to tell... they were all submitted at the same time but for manual results that's normal.

Since each curve has it's own random starting point, I wonder if a good way to avoid duplicate submissions would be to have Prime95 do a hash of the different ones for each curve and include that in the result. Then the Prime95 checksum would take that into account, which should result in unique results for each set of curves?

I'm probably not saying anything nobody else has ever thought of, so I'll retreat back into my corner now.

[Dubslow]

Originally Posted by guido72

I know that. My question was about how Primenet may distinguish from two (or more) different results if all the parameters (number of curves, boundaries, ...) it gets are the same... Or it knows that each curve is unic so that each result is unic as well...

No, PrimeNet unfortunately does not record the sigma for any given curve. (sigma is the standard variable in the literature that uniquely defines each curve.)

[retina]

And it is not just ECM that "suffers" from this problem. TF with a negative result is a trust-only thing also. We can't possibly know if any curves/TF were ever run.when the user reports no factors without doing all the work again on another system.

Fortunately the credits earned for no-factor results have no real world meaning so it really doesn't matter too much, the incentive the cheat is very low. More likely it is just a misconfiguration or misunderstanding. Except for perhaps a few extra LL/DC tests that could have been avoided no real harm is caused.

[ET_]

Originally Posted by guido72

Is this the same result or not?...

No, it's not. It was a bunch of 100 curves done on different PCs and submitted all together. The sigma values should be all different.

Unfortuntely, at the moment there is no way for Primenet to tell it from a forged result.

Luigi

[guido72]

Originally Posted by VBCurtis

Primenet has no way to know the user isn't submitting fraudulent curve counts.

I'm more than sure there's nothing fraudolent! The user is a friend of mine and an absolutely fair and reliable partecipant!!!
I was just wondering... I'm starting right now to look at ECM factoring... Never done that job before...
Regards

[guido72]

Originally Posted by retina

And it is not just ECM that "suffers" from this problem. TF with a negative result is a trust-only thing also. We can't possibly know if any curves/TF were ever run.when the user reports no factors without doing all the work again on another system.

Fortunately the credits earned for no-factor results have no real world meaning so it really doesn't matter too much, the incentive the cheat is very low. More likely it is just a misconfiguration or misunderstanding. Except for perhaps a few extra LL/DC tests that could have been avoided no real harm is caused.

Do you mean that if one just cuts and pastes in result.txt of MfaktX the usual line "NO FACTOR FOR MXXXXXXX FROM 2^xx TO 2^xx [MFAKTX 0.20 BARRETT76_MUL32_GS] just after havin' changed the exponent there is not any way to know if the test has been actually run?

[guido72]

Originally Posted by ET_

Luigi

Ciao Luigi! Tutto OK?
Salutoni!!