mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   FactorDB (https://www.mersenneforum.org/forumdisplay.php?f=94)
-   -   Other Factordb Problems (https://www.mersenneforum.org/showthread.php?t=16849)

Stargate38 2014-06-13 19:34

I'm having a problem with the Certificates page. It won't keep the "Show Unprocessed Only" Checked:

[URL]http://www.factordb.com/certoverview.php?digits=300&perpage=100&skip=0&pending=on[/URL]

Stargate38 2014-06-14 17:55

I tried adding M67,108,864 (20,201,782 digits), but it said "Error: Limit of about 10,000,000 digits exceeded". If that's true, then why is it possible to add M57885161 (17,425,170 digits) and all other known primes >10[sup]10[sup]7[/sup][/sup]?

Here's links for proof:
[URL]http://www.factordb.com/index.php?query=M57885161[/URL] (Doesn't give Max. digits error), alt link: [URL]http://www.factordb.com/index.php?id=1100000000583604171[/URL]
[URL]http://www.factordb.com/index.php?query=M67108864[/URL] (Gives Max. digits error)
[URL]http://www.factordb.com/index.php?query=M66438561[/URL] (Largest Mersenne number <10[sup]20000000[/sup], still gives error)

wblipp 2014-06-14 21:05

Perhaps Syd recently added a certificate size limitation. I had a recent email from Syd explaining that the recent stall in processing primality certificates was because of some large certificates that took over a week to process.

danaj 2014-06-14 23:52

Over a week, ouch. For a 18689 digit prime, currently the 12th largest on the primes page, I got a time of 63 hours with my single threaded program on 4770K that had 4 other processes running. Primo for the same cert took 6 hours on an idle 3930K given 12 threads. Clearly a 4770K is faster than the i7-2600, but that's pretty long.

I'd recommend upgrading to GMP 6.0.0a, as that had some pretty major speed updates (~1.2x faster) that impact these operations . Tthe above time was not with it.

Another possibility is to split up the verification into steps. E.g. a "--step <n>" option that says just do that step, and a "--structure" option to verify the structure and return the number of steps. It would allow finer granularity and not lock up one process on a very long running task.

Or maybe Syd has already found a different solution, as the status page looks like things are moving along.

chris2be8 2014-06-15 16:00

Perhaps the limit of "about 10,000,000 digits" is actually between 17,425,170 and 20,201,782 digits. Or Syd might have added M48 as a special case.

Chris

Puzzle-Peter 2014-06-15 17:07

I don't think a single huge certificate is a problem. I and others have uploaded the occasional one and though it takes a long time, there was usually one of these and two or three other workers happily processing the small ones.

Last week however, user msc_nbg uploaded hundreds of certificates for numbers with 10,000+ digits. They all differ in a tiny number of digits, so it MIGHT be the steps that the ECPP proof takes as it is created by a row of primes decreasing in size. THIS IS JUST A GUESS! Anyway this would easily clog up the certificate processing and it would be a rather easy thing to generate these certificates after having done a big PRIMO run.

Just my 2 cents and I might be completely wrong.

Mini-Geek 2014-06-15 18:20

[QUOTE=Puzzle-Peter;375901]Last week however, user msc_nbg uploaded hundreds of certificates for numbers with 10,000+ digits. They all differ in a tiny number of digits, so it MIGHT be the steps that the ECPP proof takes as it is created by a row of primes decreasing in size. THIS IS JUST A GUESS! Anyway this would easily clog up the certificate processing and it would be a rather easy thing to generate these certificates after having done a big PRIMO run.

Just my 2 cents and I might be completely wrong.[/QUOTE]

Maybe the person's thinking was, "My ECPP proof has this chain of things that are all proved prime (even though only the first was noteworthy enough to want to prove). I should tell the DB about all of them!".
Maybe the DB should (if it doesn't already) look at the certificate, and add all of the primes proved in this chain to the DB. This would also mean that it doesn't have to reprocess a certificate chain over and over. (this could be done regardless of if your guess here is right, and that's what happened here)

Puzzle-Peter 2014-06-16 09:46

[QUOTE=Mini-Geek;375906]Maybe the person's thinking was, "My ECPP proof has this chain of things that are all proved prime (even though only the first was noteworthy enough to want to prove). I should tell the DB about all of them!".
Maybe the DB should (if it doesn't already) look at the certificate, and add all of the primes proved in this chain to the DB. This would also mean that it doesn't have to reprocess a certificate chain over and over. (this could be done regardless of if your guess here is right, and that's what happened here)[/QUOTE]

After one of my first PRIMO certificates I did wonder about that myself. I was simply too lazy to take action. Plus - as others pointed out - the probability of "recycling" one of those primes is just about zero so I didn't bother.

Still they are proven primes and why not store them in FactorDB. I think extracting them from the certificates is a good idea and shouldn't be too hard to do.

wblipp 2014-06-16 16:13

[QUOTE=Puzzle-Peter;375947]the probability of "recycling" one of those primes is just about zero[/QUOTE]

I still find this argument compelling. It might be possible to convince me to support a database modification that stored the one certificate and tagged all the primes as proven by that one certificate. But I oppose the successive shortening of the certificate and resubmission after each shortening as a waste of resources - both computing and storage resources. I doubt Syd has the time to make such a database modification.

danaj 2014-06-17 01:21

It's come up before, and I wrote a simple Perl program to create new verifiable certs just to make sure it could be practically done. But I think it's a really bad idea. If desired, this is something the database should efficiently store (just a note that the proof for number N is step 45 of entry XXXXX), and generate the small cert from the larger one if requested.

Stargate38 2014-06-17 16:00

I think I found the DB's max value. It's somewhere between M62000000 and M62000001, but I was able to get M62000001 by typing in "2*M62000000+1".


All times are UTC. The time now is 08:05.

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