As of now, quite many results are being resent as a third copy (because BOINC returns an empty file). When all problems with LLR are nailed down, I'll reconsider the quorum, but for now we'll be sending two copies.

But in my oppinion, missing at least one prime is waste of processing time.

[hmm, retrying post, as the previous try returned database error]

[QUOTE=Rytis;92821]As of now, quite many results are being resent as a third copy (because BOINC returns an empty file). When all problems with LLR are nailed down, I'll reconsider the quorum, but for now we'll be sending two copies.

But in my oppinion, missing at least one prime is waste of processing time.

[hmm, retrying post, as the previous try returned database error][/QUOTE]

I agree with you my friend, since LLR isn't fail profe, and at least for me since my computer previously returned a fake prime every 1 in 2000 candidates tested, that is primes being thought to be primes, when they in fact is LLRs with residues :lol: Actually I haven't had a single error while using PrimeGrid and the LLR work that you produce, I think it is because what ever is overheating get a chance to blow out and catch its breath in the approximately 1 second there goes from finish of 1 WU to start of the next :geek:

So in short terms, I think we should stay with doing double checks, because someday the opposite situation may come, that we see primes being submitted by a user as having an LLR residue when in fact it is a twin prime :)

Or we could do a third thing in stead: "We could do only one test with LLR, and then when we find the twin prime, and while sieving is progressing, we could do double cheks of the entire work area, and see if the newly sent WU working as a doublecheck, will raise the same result or a different!" I'm not sure if this is a plausibel solution, if even possible without interfering with the internal structure --> I would actually also prefer things to be like they are for now "1 WU = 2 results (at least)"!

Btw folks, the chance that you'll recieve a PrimeGen WU when accepting to run test apps, is rather small, so there is no reason to be on the back of rytis for maybe doing two or maximum 5 PrimeGen WU a day, its not really a big of a deal. Also I think, because the results can be cheated with, that the quorum of 2 results should remain untill at least we have a way of progressing without possibility of cheating :grin:

Sorry for the rather long reply, hope it didn't act inconvenient to you :rolleyes:

Regards!

Who will get the credit if a primegrid user find a twin prime ?
Maybe we have to discuss again this question.

[QUOTE=pacionet;92826]Who will get the credit if a primegrid user find a twin prime ?
Maybe we have to discuss again this question.[/QUOTE]

KEP suggest following: Primegrid for being the biggest contributor, the 2 users finding the twin prime, the greatest siever, and (if not primegrid) the person who does most testing and find most primes, as well as those previously suggested credit divisions :rolleyes:

Btw I had a chat with Rytis yesterday, he hopes that PrimeGrid will find at least 100 primes before the end of 2006, and based on the progress from yesterday (about 10-12 M range tests) I think it is possible, so people if you wanna do double checks, please go to PrimeGrid, otherwise, of course feel free to reserve your own ranges on the TPS webpage.

Otherwise, my final remark today will be:

"Great project, with a lot of potential to be bigger than the biggest"

Regards!

If the goal of the project is to "find a record twin prime" then there should be no double checking. You will find your record twin prime in almost half the time if you do not do double checks.

If the goal is to find the "find the [I]smallest[/I] record twin prime for the n value you are testing in k*2^n+/-1" then you must do double checking to make sure you do not miss the key k value.

I would argue that the second goal is of no interest to mathematicians, so I would say the first goal makes the most sense. Thus, I recommend no double-checking.

Of course, doing some double-checks while you are making sure the BOINC platform is working properly makes perfect sense.

George, I agree with you but I should say that cheating should be also detected. If a significant amount of results would be fake, there isn't any sense at all. *Some* results could be double-checked, say random 5% to detect faulty machines and mass-cheaters.

Offtopic: I looked your avatar and noticed that the font has become bigger. After looking for more than 3 sec I thought, "hey, it is moving!" As it is 2 AM here, I thought I should have some sleep. I looked away from the monitor, then again at your avatar and understood that is was really moving.

[QUOTE=pacionet;92826]Who will get the credit if a primegrid user find a twin prime ?
Maybe we have to discuss again this question.[/QUOTE]
1.) The primegrid user, obviously :smile:
2.) Rytis, because he started primegird in the first place and allowed LLR to be one of its projects.
3.) Gribozavr, for providing us with the sieve files
4.) The user who has tested the most results overall (including among the non-primegrid ranges)
5.) The user who has found the most non-twin primes.
6.) If I'm in the top 1% of top producers (in terms of most results), then I'll be greedy and claim the credit too :devil:

---------------------

In regards to double checking, I agree with thommy, smh, and George: almost all results shouldn't be doublechecked. Only a small percentage (5-10%?) of results should be doublechecked to prevent cheating. After a user has got 3 incorrect results within a month, they should be told to check the reliability of his/her machine. More than 10 incorrect results within a month causes the user to be banned from the project for a week. If the user continues submitting many incorrect results after the ban is over, the ban should become a month, and if it happens a third time, it should become permanent.

And yes, the project's goal is to find [b]any[/b] twin prime of record size, [b]not[/b] to find the [b]smallest[/b] twin prime for n=195000.

I've got some questions for Rytis.

What is the amount of bad LLR results?
Are there users which have a significant amount of bad LLR results? (They are not cheaters, their machine can just be not stable enough)

Fail rate is 0.2874% :)

I will be going to switch random doublechecking (10% doublechecked, 90% accepted as-is), maybe this evening. However, if there is an empty file (that's the biggest problem, these results don't come in as an error) I will still resend the workunit, as I do with all error workunits.

Are all residuals returned with bonic?

I agree that double checking makes no sense. If people are cheating some parts of the range is not being checked. Anyway, this is not a problem, because sieve does not tepend on the size of the range and we have still enough left to test.

In order to find contributors who are cheating, we can simply check the number of non-twin primes (the frequency of these depends on the sieving level). If there are no or statistically many too few primes from a contributor, we coude do some sample doublechecks. Those people who are cheating want earn a lot of credit and are therefore 'testing' a large range. Inside a large range, the number of primes gets more significant and cheating is easy to detect.

Rytis , when you are near to complete all your ranges, please tell us, such that we can provide more pre-sieved ranges in time.

Each new pre-sieved release is 100M: if you need more, just tell us.