mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > PrimeNet

Reply
 
Thread Tools
Old 2009-09-07, 13:46   #12
Mini-Geek
Account Deleted
 
Mini-Geek's Avatar
 
"Tim Sorbera"
Aug 2006
San Antonio, TX USA

17·251 Posts
Default

Quote:
Originally Posted by markr View Post
Most of them have only two 64-bit residues. Early tests produced shorter residues...

Exponent,User name,Computer name,Residue,Date found
50087,Unknown,,6421,
50087,George Woltman,,7F0C06BD9A0B6421,
50087,Brian J. Beesley,cnsj,7F0C06BD9A0B6421,
True, and perhaps he was referring to this, (in which case, yes these triple-checks were useful, even just for verifying that the number is really composite, but no, further triple-checks would not be needed) but also there are some, like 100069, that really do have many matching 64-bit residues. The only way I can explain that is people playing around, and/or being unaware of the previous LLs and checking it before they run ECM curves on it.
Mini-Geek is offline   Reply With Quote
Old 2009-09-07, 13:59   #13
Brian-E
 
Brian-E's Avatar
 
"Brian"
Jul 2007
The Netherlands

2×23×71 Posts
Default

What I found with my limited number of samples of pseudorandom ranges of small (<1M) exponents was that all numbers were 3-times LL tested: many, and indeed all of the very smallest numbers, had indeed a 16-bit residue as their first test, but those that did not had - without exception in the ranges I looked at - three 64-bit residue tests. That suggested to me that someone had decided that all the exponents should have 3 tests done, allowing the original 16-bit residue result by "unknown" to be the first of the 3 if that was present, but making sure that there were 3 64-bit residue tests in cases where it wasn't.

Last fiddled with by Brian-E on 2009-09-07 at 14:15 Reason: clarification
Brian-E is offline   Reply With Quote
Old 2009-09-07, 20:59   #14
lfm
 
lfm's Avatar
 
Jul 2006
Calgary

6518 Posts
Default

Occasional triple checks might be useful in detecting fraudulent entries?
lfm is offline   Reply With Quote
Old 2009-09-07, 21:06   #15
Mini-Geek
Account Deleted
 
Mini-Geek's Avatar
 
"Tim Sorbera"
Aug 2006
San Antonio, TX USA

17·251 Posts
Default

Quote:
Originally Posted by lfm View Post
Occasional triple checks might be useful in detecting fraudulent entries?
I don't see them being very useful at that. Say a person (whether one user account or two) intentionally/fraudulently submits two identical incorrect LL results, and a triple-check shows a different residue. Won't it just be assumed that the triple check is the incorrect one? The only way you'd detect the fraud is if an unusually high percentage of the fraudulent user's accounts are recognized as having unmatching triple checks.
Mini-Geek is offline   Reply With Quote
Old 2009-09-08, 00:51   #16
cheesehead
 
cheesehead's Avatar
 
"Richard B. Woods"
Aug 2002
Wisconsin USA

22·3·641 Posts
Default

Quote:
Originally Posted by Brian-E View Post
Can anyone tell us why so much of the lower end of the exponents have been triple checked?
Several years ago, Brian Beesley, a great early contributor to GIMPS, performed a deliberate and systematic run of LLs through all exponents up past 220, to ensure that each had at least two 64-bit matching residues. At that time, many of those exponents had only one 15-bit residue (from from David Slowinski) and one 64-bit residue.

Beesley didn't have afforable Net access, so communicated only by e-mail.

Quote:
Is it just people having fun, is it for the error rate data, or are there other concrete reasons too?
The latter. It provided data for checking the validity of our group deduction/assumption that two matching residues process provides all the assurance we need.

It's one thing to calculate. It's another to go out and do actual measurements. :-) So, someone did it.

Quote:
Originally Posted by Brian-E View Post
Presumably the intention is for that to be done further?
No, unless someone else volunteers. Beesley stated that he was stopping there, with no intention of continuing.

Quote:
Originally Posted by Brian-E View Post
the only serious use of triple checking would seem to be with completely different client software to allow for the possibility of a bug as you mention, and that is largely handled by the random offset at the start of a LL test, we would hope.
I don't recall what software Beesley used. He was well aware of the need to have his triple-checks as independent as possible of previous results. I think it's safe to assume that whatever he used did provide that independence as far as he could manage.

- - -

Quote:
Originally Posted by Mini-Geek View Post
I'm 99% sure that triple checks are very close to being totally useless. (unless for some other purpose...definitely not useful in the same way that double checks are useful)
But triple checks are useful in the same way that double checks are! It's just that the percentages are more extreme.

Quote:
I think the odds of Prime95 having a bug that would produce matching residues (which is very small) is better than the chance that two 64-bit residues just happened to match when they're both wrong, so a third check wouldn't be useful.
Well, there's thought ... and on the other hand there's actual experiment. The latter trumps the former.

Aristotle thought a heavy object would necessarily always fall faster than a light object. A millenium-and-a-half later, Galileo demonstrated otherwise.

Quote:
The extra LL results (triple-, quadruple-, etc. checks) in small exponents is likely due to people playing around. (one time, I went through a list of small-ish primes and re-discovered the first couple dozen or so Mersenne primes, just for fun if I had submitted my results to PrimeNet, they'd appear as more matching LL results on those small numbers)
Not everyone has only those same motives. :-)

Quote:
Originally Posted by Mini-Geek View Post
also there are some, like 100069, that really do have many matching 64-bit residues. The only way I can explain that is people playing around, and/or being unaware of the previous LLs and checking it before they run ECM curves on it.
There are other explanations besides those.

Sometimes a software glitch caused systems to repeatedly test and report on the same exponent.

Sometimes an erroneous setup in local distribution resulted in having multiple (usually, remote) systems run and report the same first-time test.

Last fiddled with by cheesehead on 2009-09-08 at 00:54
cheesehead is offline   Reply With Quote
Old 2009-09-08, 18:16   #17
Mini-Geek
Account Deleted
 
Mini-Geek's Avatar
 
"Tim Sorbera"
Aug 2006
San Antonio, TX USA

17·251 Posts
Default

Quote:
Originally Posted by Mini-Geek View Post
I've got an idea: when a matching DC is returned on a candidate that is already assigned, set a flag on that assignment so that the next time Prime95 communicates the status on that assignment, PrimeNet tells it to stop and report its percent completion, then give partial credit based on that percentage. (e.g. if you're at 90%, and the credit would be 14, give 14*.9=12.6) (or, if it's reporting that it's complete, check that the reported residue matches the other two residues and give full credit, or whatever it normally does if you report a result that ends up not matching)
Mini-Geek is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Posts that seem less than useless, or something like that jasong Forum Feedback 1050 2019-04-29 00:50
Fedora gedit for bash has become useless EdH Linux 11 2016-05-13 15:36
Useless SSE instructions __HRB__ Programming 41 2012-07-07 17:43
Useless p-1 work jocelynl Data 4 2004-11-28 13:28

All times are UTC. The time now is 22:28.

Wed Aug 5 22:28:48 UTC 2020 up 19 days, 18:15, 2 users, load averages: 1.01, 1.28, 1.44

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.