mfaktc: a CUDA program for Mersenne prefactoring

[James Heinrich]

Originally Posted by drh

Is this what is expected?

Yes (for now). PrimeNet will give a message to the effect of "insufficient information to correctly determine credit" and take its best guess (based on the size of the factor and exponent) as to whether you most likely found the factor using TF, P-1 or ECM.

It would probably be fairly trivial to rewrite the processing code to assume that if it's a mfaktc result line then it's always a TF result, but PrimeNet does not currently make any such assumptions (so it's wrong about how the factor was found, more often than not, with mfaktc results). If it guesses wrong, then CPU credit is almost certainly wrong, but I've seen it wrong both ways (too high and too low).

[Christenson]

Originally Posted by James Heinrich

Have you put any thought into how appropriate work quantity and range is determined for a given system? The number of "at least 30 minutes per assignment" was thrown around, but there will be systems pushing through anywhere from 25 to 500+ GHz-days/day, and what's an appropriate assignment for one system might not be for another. Would you make an attempt to examine the throughput rate of mfaktc and determine appropriate work levels that way, or rely on the user editing the .ini file to say "my system can do ____ GHz-days of work per hour"?

James:
I was thinking about this problem from a server perspective; that is, if there are large numbers of systems out there that contact the server every minute, it will look like a continuing DoS attack on the server.

I looked at the code last night, and I'll say that Oliver seems much more interested in the CUDA side than the silly administrative details, so that part of the code is quick-to-write and straightforward. But it makes having the main thread simply ask if enough wall-clock time has elapsed since the beginning of the factoring range or do another bit level straightforward.

Therefore, I think there should be 2 keys:
One to automatically increase the bit level by the amount desired, say, 3 steps.
One to specify a minimum wall-clock time for a given exponent.

E.

[TheJudger]

Originally Posted by Christenson

I looked at the code last night, and I'll say that Oliver seems much more interested in the CUDA side than the silly administrative details

This is not a secret. Keep in mind that mfaktc was intended as a proof-of-concept.

I'll take a look at your draft specification later!
One detail we've to think about are the licenses of the mprime-code and mfaktc (GPLv3). Opensource doesn't mean that you can do whatever you want.

Oliver

[Christenson]

Originally Posted by TheJudger

This is not a secret. Keep in mind that mfaktc was intended as a proof-of-concept.

I'll take a look at your draft specification later!
One detail we've to think about are the licenses of the mprime-code and mfaktc (GPLv3). Opensource doesn't mean that you can do whatever you want.

Oliver

If mfaktc is your idea of a proof-of-concept, we'd better watch out for when you do stuff for real!!!!

I need to update my spec with a list of the new mfaktc.ini keys.

Copyright wise, I'm doing this for bragging rights and an unbounded negative crank score here. List me as a contributor, teach me some more math, tolerate my sillier ideas (and shoot them down with good reasoning), I'll be happy.

I'm assuming GPL v3 means I can "kitbash" the codes together and release the results under GPL v3. In any case, I somehow don't think prime95 will mind having prime95 code show up in mfaktc. It's not like I'm going to copy the hard parts, which are the ones that do the math.

[Prime95]

Originally Posted by Christenson

If mfaktc is your idea of a proof-of-concept, we'd better watch out for when you do stuff for real!!!!

I need to update my spec with a list of the new mfaktc.ini keys.

Copyright wise, I'm doing this for bragging rights and an unbounded negative crank score here. List me as a contributor, teach me some more math, tolerate my sillier ideas (and shoot them down with good reasoning), I'll be happy.

I'm assuming GPL v3 means I can "kitbash" the codes together and release the results under GPL v3. In any case, I somehow don't think prime95 will mind having prime95 code show up in mfaktc. It's not like I'm going to copy the hard parts, which are the ones that do the math.

If you put any code under GPL v3 it will never be incorporated into prime95.

[James Heinrich]

Originally Posted by Prime95

If you put any code under GPL v3 it will never be incorporated into prime95.

Unless it's dual-licensed under a Prime95-compatible license (what license is Prime95 code licensed under?)

[Prime95]

Originally Posted by James Heinrich

Unless it's dual-licensed under a Prime95-compatible license (what license is Prime95 code licensed under?)

It is open source with 2 exceptions:

1) The security code that makes it a tiny bit harder to forge results is not public.
2) You cannot use the code to Mersenne primes unless you agree to abide by the GIMPS prize rules.

The first restriction will preclude you from releasing under GPL.

[Christenson]

1) Here is my latest spec. I've decided which new keys will be essential, and which optional.

2) mfaktc seems to be under GPL v3. I want to a) use the public parts of prime95 to save effort extending and improving mfaktc, b) add a little glue code of my own, with whatever license and copyright that makes sense, and c) see my name in public places. (like it hangs out now near the top of the ECM list). I get paid back as I learn to do things like using the eclipse development environment, make Linux boxes work, etc. For an encore, I'd like to do a CUDA P-1, but that's too ambitious right now, and I did say I wanted to work on Minix, where I think some real impact can be made with the security model.

I recognize the necessity of legalities, but honestly, I want to fool with code, not lawyers, and maybe make a name for myself with M49. From a technical standpoint, I would like to have the security module incorporated into mfaktc, although I don't know how that would work or whether such measures would have any hope of being effective.

Anyway, I have LL results due shortly. Oliver and George, please work out the correct legal basis. Upgraded specification attached.

[Prime95]

I am not a lawyer.

I don't know if you can "undo" GPL on mfaktc and release it as public domain instead.

[xilman]

Originally Posted by Prime95

I am not a lawyer.

I don't know if you can "undo" GPL on mfaktc and release it as public domain instead.

IANAL as well. However, I have studied intellectual property legislation as it applies in a number of jurisdictions and in some detail.

In summary, the copyright owner has complete control over the intellectual property concerned and can release, or not, anything they wish under any conditions they wish.

In particular, you could be granted a personal license to use a modified version of the sources in a manner which you and the owner find mutually acceptable. The modification could be as simple as replacing all the copyright notices and GPL license texts in the sources with a mutually acceptable variant.


Paul

[chalsall]

Further to xilman's above, you (George and Christenson) should be aware that Christenson cannot use George's code and release it as GPL without George's permission to do so.

This assumes that George's code is not Public Domain. Based on the fact that there are copyright notices ("All rights reserved.") in the source files, the code is not Public Domain.

And xilman is correct that the original author(s) have the right to re-release the code under any license they want, but please note that the originally released code remains under the original license (even if they're the exact same).