Msieve with GNFS support

[_dc_]

Actually, the software-engineering considerations are secondary to the need to take the time and understand the computational algebraic number theory behind how the square root works, and I've tried to do that with very little success.

Believe me, I know what it takes trying to understand some kick**s maths
I'm now finishing CS department and for the first three years we were ony studying maths, all of its sorts and flavours.

The easiest way to run the square root in parallel is to have different machines run different dependencies, after putting the relation file in a shared network directory. You don't want a single machine to process multiple dependencies because of the memory use involved.

Yep, I understand this, and this is what I was trying to do. The problem was that another machine wasn't connected to network so copying all files using DVD-RW took some time.

What about putting .exe built with /LARGEADDRESSAWARE on your web site? This would help people with Windows environment.

Originally Posted by R.D. Silverman

All one would have to do is convert the (output of the) msieve matrix
solution to the CWI format and convert the relation format as well.
I believe code exists for the latter.

Studying formats and coding converters could be tricky sometimes. I don't know exactly about this case, but I suppose it will take me more time than computing few square roots (as I'm not familiar with Msieve/CWI source).

[jasonp]

Originally Posted by _dc_

What about putting .exe built with /LARGEADDRESSAWARE on your web site? This would help people with Windows environment.

Sure, it turns out this is easy (I didn't know).

Studying formats and coding converters could be tricky sometimes. I don't know exactly about this case, but I suppose it will take me more time than computing few square roots (as I'm not familiar with Msieve/CWI source).

There are programs that can convert relations both ways. Nobody has (yet) written programs that convert the formats of sets-of-relations that both suites need

[frmky]

Originally Posted by R.D. Silverman

All one would have to do is convert the (output of the) msieve matrix
solution to the CWI format and convert the relation format as well.
I believe code exists for the latter.

The hard part is obtaining a license to use the CWI code. In my experience, they aren't very generous.

Greg

[xilman]

Originally Posted by frmky

The hard part is obtaining a license to use the CWI code. In my experience, they aren't very generous.

In my experience, they are very generous. I acquired a personal license for myself and, in separate transactions, licenses for the other NFSNET team members without any fuss at all.

It helps enormously if you can show a track record in factoring large integers --- that you're reasonably knowledgeable, are unlikely to ask for help and have the tenacity to stick with long-term projects.

Of course, my experience is several years old and the situation may have changed since then.

Paul

[R.D. Silverman]

Originally Posted by xilman

In my experience, they are very generous. I acquired a personal license for myself and, in separate transactions, licenses for the other NFSNET team members without any fuss at all.

It helps enormously if you can show a track record in factoring large integers --- that you're reasonably knowledgeable, are unlikely to ask for help and have the tenacity to stick with long-term projects.

Of course, my experience is several years old and the situation may have changed since then.

Paul

In my experience, they have also been quite generous.

[frmky]

Originally Posted by xilman

It helps enormously if you can show a track record in factoring large integers --- that you're reasonably knowledgeable, are unlikely to ask for help and have the tenacity to stick with long-term projects.

That's the problem. Many people here are new to factoring and thus don't have a long track record. I'm not surprised that they have been generous toward both Bob and NFSNet. Further, having now been "adopted" by NFSNet I would probably be able to get a license easily as well, but as an unknown in the field, even with an academic title, my requests were originally ignored, then rejected.

Greg

[xilman]

Originally Posted by frmky

That's the problem. Many people here are new to factoring and thus don't have a long track record.

It may be a problem, but it's a solvable problem. Bob and I were newbies once.


Paul

[bdodson]

Originally Posted by xilman

It helps enormously if you can show a track record in factoring large integers --- that you're reasonably knowledgeable, are unlikely to ask for help and have the tenacity to stick with long-term projects. ...Paul

In my case, the large number was a share of the sieving for RSA140,
and some sieving experiments with Peter shortly after. A complete suite of
sgi binaries, along with lots of explanations and suggestions from Peter.
Things changed after RSA155=RSA512, when CWI signed with MS to
license the suite. Paul went to MSR. Peter to MS. There often seemed
to be a question of what would be said by the relevant lawyers, either
CWI's, MS's, or both. I still got updates on the filter, including Steffi's
code; but not much after SNFS768 (2,773+) and the first parallel version of
block Lanczos (for which even large amounts of begging didn't suffice for
a copy of the sgi binary). We switched over to commercial standards;
academic standards no longer applied. Not that academics have an
especially spotless record on sharing.

Incidently (or not), we've just passed a milestone with Tom's factorization
of 7^387-1 C154 and jbristow's 5,775M C151. The new smaller-but-needed
Cunningham's will be at/over 512-bits! -Bruce

[jasonp]

As a sneak peek into the changes for v1.34, I've updated the NFS filtering to behave more like what I imagine the CWI suite does. Instead of repeating the filtering with smaller and smaller bounds on large ideals, the attached will perform two passes only, and the seond pass has an extremely small bound. The relations are initially written to disk, and only ideals that occur in less than X relations are read back into memory.

This should make the filtering a good deal faster, yield matrices that are a few percent smaller, and most importantly should make NFS filtering a lot more consistent. Anyway, that's the plan; I've tested it on a few examples and it works pretty well. What I wonder is if the new code does a better job than v1.33 on datasets you have lying around, especially datasets that are heavily oversieved. If you can give the new code a try, let me know how it goes (please PM if possible).

Thanks,
jasonp

[BotXXX]

Originally Posted by frmky

That's the problem. Many people here are new to factoring and thus don't have a long track record. I'm not surprised that they have been generous toward both Bob and NFSNet. Further, having now been "adopted" by NFSNet I would probably be able to get a license easily as well, but as an unknown in the field, even with an academic title, my requests were originally ignored, then rejected.
Greg

As a side note - i also didn't have any problem with obtaining a license for the CWI suite. Just contacted the CWI desk and later received a reply with an agreement to use the suite. (and I don't really have any big/long track-record in this area) (around end of 2004 i think)

[henryzz]

if anyone wants old versions of msieve ever i just found these
http://sage.math.washington.edu/home/wbhart/SIMPQS/