GIMPS and bitcoin

[Rodrigo]

GIMPS recently got a little plug in a totally unrelated forum.

Aside from that, how about the idea of awarding bitcoin for finding a Mersenne prime? (Assuming that the digital currency concept doesn't collapse as a result of bitcoin's recent troubles.) It would somehow be fitting to award bits for processing bits. TBH, though, I'm not sure that in the end I wouldn't rather have cold, hard cash (OK, a check) in my hands.

Rodrigo

[f1pokerspeed]

The problem with BTC is that the value changes significantly from day to day. The currency has been so volatile that it is impractical for use as a reward.

[TheMawn]

There was a full-on discussion of MersenneCoin not too long ago.

In summary, the value is way too volatile, and awarding it would be even harder. Any award would have to be double-checked before handing it out. Maybe factors found would award coins but that would only create a feeding frenzy for the 100M-1000M TF range and wouldn't provide any meaningful, useful work.

[kladner]

DANGER Will Robinson!

I feel that adding something like that would dratically change the scenery, as it were. It would introduce competition beside which everyday GHz-d wh0ring pales, even though the perception of value of this currency might, at any time, prove specious.

[c10ck3r]

Originally Posted by TheMawn

Maybe factors found would award coins but that would only create a feeding frenzy for the 100M-1000M TF range and wouldn't provide any meaningful, useful work.

Because, you know, we will NEVER LL test over n=100M...

Oh, wait.

[Mini-Geek]

Originally Posted by c10ck3r

Because, you know, we will NEVER LL test over n=100M...

Oh, wait.

I think it could be argued that, given the advent of GPUs for TF, all CPU pre-TFing (that is, TF that was not followed by a pre-GPU LL test) provided nearly no meaningful, useful work. We couldn't really have known that ahead of time, but it's still the case.

Originally Posted by TheMawn

Maybe factors found would award coins but that would only create a feeding frenzy for the 100M-1000M TF range and wouldn't provide any meaningful, useful work.

Not if you award coins by the GHz-days credit, instead of a fixed amount per factor. But it would give incentive to TF far more than is optimal (people other than Mersenne enthusiasts would try to optimize earnings, not prime finding rate), and deciding how to credit the coins could be tricky. E.g. you might find factors via P-1 and try to get them reported as TF, or vice versa, in order to maximize credit (we have enough trouble identifying them as-is). You'd also have to think about how PrimeNet gives different credit for factors in different portions of the bit range (because Prime95 searches in a certain order). That should probably just be nixed (for coin credit purposes) in such a scenario, since otherwise you could search only the highest-credit portions, and not clear out factors for LL purposes.

There's another problem with using Mersenne factors as a proof-of-work: typical proof-of-work requires that you have new work for each block, so that you can't e.g. find 10 factors beforehand, and then release 10 blocks at once based on those factors. You'd have to have a way to give limit it appropriately or use a different block generation scheme while still giving coins for factors.

[kladner]

Here there be Tygers.

http://en.wikipedia.org/wiki/Here_There_Be_Tygers

[TheMawn]

Originally Posted by kladner

I feel that adding something like that would dratically change the scenery, as it were. It would introduce competition beside which everyday GHz-d wh0ring pales, even though the perception of value of this currency might, at any time, prove specious.

The big idea is to attract more computing to the project by giving a cash incentive. I don't think the aspect of competition would be a problem. It would definitely change the environment around here. Some spoiled kid with a big inheritance (or something) could come along with his 15 Titan Blacks, climb to the top-ten in a month and walk around like some kind of big cheese, and yeah, that would be annoying.

As it stands, we're all a bunch of volunteers. The only thing I like more than this project is what it represents: distributed computing. That's why I spent $250 on a GTX 660 Ti when I didn't need to. It has rendered maybe 30 minutes of 3D graphics and maybe an hour of desktop over the last eight months because it was bought for the express purpose of trial factoring for GIMPS. And I don't regret that decision. That's my personal contribution to the project.

The monetary gains notion will change the environment, definitely. For better or worse. There's no way to tell.

Originally Posted by c10ck3r

Because, you know, we will NEVER LL test over n=100M...

Oh, wait.

That's not where I was going. I'm not that naive. I was suggesting that monetary gains will attract new people to 100M (or worse, divert current work from to 100M TF) which is meaningless work in that at our current pace it will only be helpful within maybe five years, more if 1/4 of our firepower leaves the wavefront.

The long story is you can only give credit for double-checkable work. You can't credit "No Factor" work because it's obscenely easy to fake. You can credit "Factor Found" but it creates an incentive to work on the lowest available bits. Is a 550M factor found valuable in the long (long long long long) run? Yes. Is it valuable now? No. We're short on trial factoring power at the wavefront as it is. Moving it over to more lucrative lands is not helpful.

I see trial factoring as a way to remove LL candidates. In my eyes, prime-or-not is the meaningful work, not "No factor for M8xx,xxx,xxx from 2⁶⁶ to 2⁶⁷".

Originally Posted by Mini-Geek

Not if you award coins by the GHz-days credit, instead of a fixed amount per factor. But it would give incentive to TF far more than is optimal (people other than Mersenne enthusiasts would try to optimize earnings, not prime finding rate), and deciding how to credit the coins could be tricky. E.g. you might find factors via P-1 and try to get them reported as TF, or vice versa, in order to maximize credit (we have enough trouble identifying them as-is). You'd also have to think about how PrimeNet gives different credit for factors in different portions of the bit range (because Prime95 searches in a certain order). That should probably just be nixed (for coin credit purposes) in such a scenario, since otherwise you could search only the highest-credit portions, and not clear out factors for LL purposes.

There's another problem with using Mersenne factors as a proof-of-work: typical proof-of-work requires that you have new work for each block, so that you can't e.g. find 10 factors beforehand, and then release 10 blocks at once based on those factors. You'd have to have a way to give limit it appropriately or use a different block generation scheme while still giving coins for factors.

You would have to give credits for a factor, based on perhaps the midpoint of the range. It might also have to be limited to one factor per exponent to remove incentive from (rather, not ADD incentive to) scouring the upper ranges for tiny factors. One example is a user named TJAOI who searched for factors across the entire range of exponents but doing it by class rather than by exponent, so he immediately found hundreds of thousands (he literally stands at 351,087 of 351,156 in TF) of factors, all in the 2⁴⁰ area. Do we make him a mersennellionaire?

Again, this comes back to the definition of the project. Yes, finding all factors has some finite mathematical value but the point of the project is to find mersenne primes. No amount of trial factoring achieves that, so why reward it?

[Brian-E]

This sort of discussion always makes me want to laugh nervously, but I feel like I'd better not because everyone else is being so serious. It's better just to admire the emperor's new clothes like everyone else around us, listen to the erudite opinions on which parts of the finery are the most important (yes, of course. his gorgeous hat and his perfect shoes have some finite value, but it's his beautiful flowing robe which is the whole point). And just ignore and pretend we can't see the rather large ... um.... elephant in the room.

Participation in GIMPS is actually quite a good model for life itself, I think. The whole project is wondrous, remarkable and brilliant in its conception, and a lot of fun to be involved in if you're able to make that from it. But questions about its ultimate point don't really lead to any satisfactory answer.

[f1pokerspeed]

The solution to TF is create a formula for the amount of coins you're going to give for the tasks. Ideally, to avoid any fake results, you would need to only award coins for any factors found.

I propose we use the formula



...where



and



For the sake of clarity, G_trial is the GHz-d/d value w.r.t. trial division for the user. P is the average GHz-d/d value for all users.

My reasoning for the (complicated?) formulae is that this gives a fair representation of the value of the factors found, and the effort/energy exerted to find them. Those with larger firepower will see more coins/factor at the same bitlevel as the "small fry" though, which is a problem. Maybe I should apply a (small?) penalty if G_trial / P is more than a standard deviation away... or would this balance out fairly as the "difficulty" introduced by a higher TFLOPS level for the entire project? The other problem is that if I were to account for this, I would need to know the TFLOPS distribution across the data range (which is an issue), considering that I highly doubt the data follows a normal distribution (and as such a standard deviation would be inappropriate). I shall leave you guys to figure that out (statistics & algorithms are not my thing).

[TheMawn]

Alternatively, coins can be handed out solely on work issued by the server.