Holy new Mersenne prime, Batman! (M47 related)

[Batalov]

The divide-and-conquer verification would have been very sweet, indeed.
(the residues are compatible between different programs, but the save files? who knows; at least they are not even downward-compatible between 25.x and 24.x... Not an serious problem, though, to reformat them.)

Just to spell out the obvious: Jud proposed to give each verifier process a task, starting from one interim save file to arrive to the next one and observe the match.

Anyway, not everyone uses, say, InterimFiles=1000000 (I do, but look how much good it had done for me - the disk is full of crap and all candidates are composite... Nevertheless, I still keep a couple 41M-42M-ers to eventually check how the 2306K FFT works.)

But if the lucky owner did - it's a very nice thought, Jud.

--Serge

[S485122]

Originally Posted by Jud McCranie

How about this for faster confirmations - have Prime95 save its data after 10%, 20%, 30%, etc of the iterations. Then 10 instances of a confirmation program could be run in parallel, making the wait for a confirmation much shorter.

Saving 26 million interim files to speed up 45 verifications ?

We waited for more than a year to (perhaps) find a number, and should spend so much effort and resources to get an answer 10 days earlier ?

Jacob

[Batalov]

Originally Posted by cheesehead

The icosahedral dice rolls.

Heeeeey, I have a set of those. What am I doing wrong?

[Jud McCranie]

Originally Posted by S485122

Saving 26 million interim files to speed up 45 verifications ?

We waited for more than a year to (perhaps) find a number, and should spend so much effort and resources to get an answer 10 days earlier ?

Jacob

I didn't say 26 million files, I said 10 files. Prime95 saves one set of intermediate files anyway. Hard disks are large and the files are relatively small. Save 10 files at different stages instead. If the LL test fails, delete the files. If the first LL test pases, use those files on 10 different computers to get a quick confirmation.

A quick confirmation will help keep it from leaking, cut down on the speculation, guessing or figuring out the exponent, etc.

[cheesehead]

Originally Posted by Batalov

Heeeeey, I have a set of those. What am I doing wrong?

Failing to collaborate with Mini-Geek.

[Andi47]

Originally Posted by Jud McCranie

How about this for faster confirmations - have Prime95 save its data after 10%, 20%, 30%, etc of the iterations. Then 10 instances of a confirmation program could be run in parallel, making the wait for a confirmation much shorter.

To do this sort of "divide and conquer" verification, the verifyers need not only the savefiles, but also the computing power (and I am not talking about P4's, but about "different hardware" such as Itaniums.)

AFIK, The verifications are already running multithreaded, so I guess they are using all computing power, which is available to the verifyers. If this is correct, the "divide and conquer"-thing would not speed up the verification.

[Mini-Geek]

Originally Posted by Jud McCranie

I didn't say 26 million files, I said 10 files. Prime95 saves one set of intermediate files anyway. Hard disks are large and the files are relatively small. Save 10 files at different stages instead. If the LL test fails, delete the files. If the first LL test pases, use those files on 10 different computers to get a quick confirmation.

A quick confirmation will help keep it from leaking, cut down on the speculation, guessing or figuring out the exponent, etc.

Prime95 doesn't already save one set at a fixed point. It does save two if the option is enabled, but that's just 30, 60, min ago, not half way or something. I agree that saving 10 (or so) intermediate files and then deleting them if it returns composite would be good with little negative effects, and should be enabled by default, but there still should be an option to turn it off for people that don't want to use up that hard drive space or for whatever reason don't want it on.

I'm not sure where he pulled that 26 million number from either BTW. If it was implemented and all clients were using it and had an average of 5 files (average completion time for all will always be ~50%), that'd be ~418,000 files total. At 4,883 KB each (a test on the first candidate over 40M shows that's the size at that range), that's a total of ~24 MB extra per copy of Prime95, or ~1.9 TB extra total for all GIMPS. This might be a bit high estimate due to reserved exponents that haven't started, but you get the idea. It's not a huge deal, but we should at least give an option for those very tight on hard drive space.

But yeah the questions other people brought up about cross-platform compatibility (necessary for a 100% sure double check) are quite valid. Somehow I don't think they'd want to go through the trouble of making the feature and making those files cross-platform and cross-program just to speed up the prime double checks.

[Jud McCranie]

Originally Posted by Mini-Geek

But yeah the questions other people brought up about cross-platform compatibility (necessary for a 100% sure double check) are quite valid. Somehow I don't think they'd want to go through the trouble of making the feature and making those files cross-platform and cross-program just to speed up the prime double checks.

OK. To me, though, it seems like it is just doing what it would be doing anyway. The file would say "this is the residue after 4,000,000 iterations for this exponent", "this is the resudue after 8,000,000 iterations", then a program on any platform could take the residue after 4,000,000 iterations and run 4,000,000 more iterations and see if the residue matches. Rather than starting with 0 residue and checking to see if the last one is 1, start with a residue given in a file and check the end with the other file.

Ten or more computers (different from PCs) could run them in parallel. But maybe it is more trouble than it is worth.

The files are only about 10 MB, so 10 of them would only be 100MB, and I don't you can even buy a HD smaller than 80,000 MB today.

[ixfd64]

It looks like we got mentioned on Scientific American!

http://www.sciam.com/blog/60-second-...-fo-2008-08-28

[S485122]

Originally Posted by Mini-Geek

I'm not sure where he pulled that 26 million number from either BTW.

"He" just multiplied the number of primes bellow 43000000 by ten. I did indeed chose to forget to take into account the fact that one does not need to save interim files for factored numbers and that once a number has been computed not to be prime, the interim files can be discarded.

Speeding up the verification would deprive the more placid people of the funny spectacle of all those who cannot hide their impatience once a possible prime has been reported.

Jacob

[Batalov]

Originally Posted by Jud McCranie

The files are only about 10 MB, so 10 of them would only be 100MB, and I don't you can even buy a HD smaller than 80,000 MB today.

They are even twice less now, ~5Mb (I am talking about 25.x) _and_ they are perfectly cross-platform. I only mentioned that they cannot be used in 24.x without certain reformatting (I am not talking about any details here) - but in essence, no extra programming work is needed. (it is already done by George - by making new savefiles)

Just a thought (for 25.7): make a default InterimFiles=4000000, and have it enabled by default, and add a disabling setting to the GUI, and add removal of interim files to the cleanup code (and another option to disable this cleanup, off by default).