Current status

[xilman]

Originally Posted by R.D. Silverman

The last report on the web site was for May 11.

2,797+ should have finished sieving.????

What is the current status on both numbers?

I finished sieving 2,1041+ and am doing 2,1426L.

According to mail that Richard sent to Sam Wagstaff today:

> 2,797+        c150    NFSNET -- Sieving complete
> 3,479+        c197    NFSNET -- Sieving in progress

I know no further details myself.

Paul

[Wacky]

I have just returned home and, although I have some major problems unrelated to NFSNet which require my attention, I can report that I have done some of the filtering on 2,797+ and am presently refactoring the relations to recover the smaller ideals that are not transmitted by the sievers. That task should finish tonight and I will build the trial matrix to see how many excess relations are available to reduce the matrix size.

As for 3,479+, I can report that we have completed 37% of the sieving and that all active sievers are working on that problem.

[R.D. Silverman]

Originally Posted by Wacky

I have just returned home and, although I have some major problems unrelated to NFSNet which require my attention, I can report that I have done some of the filtering on 2,797+ and am presently refactoring the relations to recover the smaller ideals that are not transmitted by the sievers. That task should finish tonight and I will build the trial matrix to see how many excess relations are available to reduce the matrix size.

As for 3,479+, I can report that we have completed 37% of the sieving and that all active sievers are working on that problem.

The last update was June 1. What is the current status?

[Wacky]

The matrix processing for 2,797+ should finish this weekend.

3,479+ is now at 49%.

[fivemack]

OK, so NFSNET is managing 12% a month, at which rate the sieving won't be done until the end of the year; 3^497+1 is a good deal smaller than numbers that NFSNET has previously completed more quickly than that, which sounds as if it's starting to run out of steam.

Getting the stats back up might help; I'd be interested in getting them working, and think I have roughly the right set of skills. Looking at other distributed-computing projects, getting some form of competition between the partisans of different processors working seems a good way to accumulate CPU-years; are we gathering the data needed to do that, and do we have the permission of our users to disseminate it?

[bdodson]

Originally Posted by fivemack

OK, so NFSNET is managing 12% a month, ... sounds as if it's starting to run out of steam.

Getting the stats back up might help; I'd be interested ... some form of competition between the partisans of different processors working seems a good way to accumulate CPU-years; are we gathering the data needed to do that, and do we have the permission of our users to disseminate it?

To the extent to which there has been a drop-off in relations/day,
two visible changes are the end of the daily Stats, and splitting the
sieving effort over two simultaneous projects (gnfs and snfs). I'll be
happy to see 2,797+ C150 completed, and we seem to have learned
some things from this gnfs experiment, but waiting a year for a project
to complete requires a longer attention span than one ought to expect
from contributors.

One comparison between NFSNET and the projects run by the cabal
is that cabal factoring announcements (RSA140, RSA155 = RSA512,
SNFS774) always featured the contributions of the people doing the
sieving, in addition to post-processing info (filtering, matrix). If there
are too many to list (but compare RSA129, where the published paper
lists the contributors!), perhaps there could be a threshold for people
to meet --- top 20, or 3% of the total, with special mention of early
adopters (preferably continuing from the previous project) that ran from
start to finish. I was certainly happy to have my relations_collected
included in the cabal announcements; but by contrast, I don't even
know what proportion of 2,1466L I did. The 2,797+ and 3,479+
reports would seem to be a good place to start!

Bruce Dodson, Math Dept, Lehigh Univ

[R.D. Silverman]

Originally Posted by bdodson

To the extent to which there has been a drop-off in relations/day,
two visible changes are the end of the daily Stats, and splitting the
sieving effort over two simultaneous projects (gnfs and snfs). I'll be
happy to see 2,797+ C150 completed, and we seem to have learned
some things from this gnfs experiment, but waiting a year for a project
to complete requires a longer attention span than one ought to expect
from contributors.

One comparison between NFSNET and the projects run by the cabal
is that cabal factoring announcements (RSA140, RSA155 = RSA512,
SNFS774) always featured the contributions of the people doing the
sieving, in addition to post-processing info (filtering, matrix). If there
are too many to list (but compare RSA129, where the published paper
lists the contributors!), perhaps there could be a threshold for people
to meet --- top 20, or 3% of the total, with special mention of early
adopters (preferably continuing from the previous project) that ran from
start to finish. I was certainly happy to have my relations_collected
included in the cabal announcements; but by contrast, I don't even
know what proportion of 2,1466L I did. The 2,797+ and 3,479+
reports would seem to be a good place to start!

Bruce Dodson, Math Dept, Lehigh Univ

Part of the dropoff (about 50%) was due to the fact that Paul
left Microsoft Labs. Active machines dropped from 200 to about 100.

NFSNET is limited by two things: available time of the people running the
project and the fact that one can't sieve offline then send in relations
in batches. The ability to do the latter would/should result in many more
sievers. However, the post-processing of each number requires people
time, and that is limited.

I, for example, am doing an NFS factorization every 3 weeks (or so).
If I had more sievers it would speed up sieving elapsed time, but I
would need more of *my* time to handle the post-processing. It takes
several hours to gather all the data and manage the filtering process.
[filtering is very hard to automate for large data sets if one wants good
matrices; it requires a lot of hand tuning]. Once filtering is done, I
just launch the LA and this only requires a few minutes of my time to manage.
[e.g. restarting if the machine goes down etc.]

The post processing is labor intensive.

Note: 2,1044+ will finish sieving in 2 days as will the LA for 2,1426L.
I am on vacation all of next week, so post-processing on 2,1044+ will wait.
I will then do, in progression, the first 5 holes in the 2LM table:
1454L, 1462L, 1478L, 1490L, and 1526L. This will finish the 2LM tables
to 768 bits.

My siever is quite a bit faster than the one used by NFSNET, but I only
have ~15% of their computing power.

[R.D. Silverman]

Originally Posted by R.D. Silverman

<snip>

What is the status of 2,797+?

[Wacky]

This one went to the 4th dependency.

Factorization completed after 32284.53 seconds, at Mon Jul 3 21:29:58 2006
Original number had 150 digits:
808327788723521825530207728143259941867725165285280602423994411926146120344539039289407533152438819052946060624026017056923683548235948384716482729353
Probable prime factor 1 has 60 digits:
721081346319803014103048601140804555897014469307122160867977
Probable prime factor 2 has 91 digits:
1120993897358460578404490502184135771743858065308635982721119473885677535372544659994092289

[R.D. Silverman]

Originally Posted by Wacky

This one went to the 4th dependency.

Factorization completed after 32284.53 seconds, at Mon Jul 3 21:29:58 2006
Original number had 150 digits:
808327788723521825530207728143259941867725165285280602423994411926146120344539039289407533152438819052946060624026017056923683548235948384716482729353
Probable prime factor 1 has 60 digits:
721081346319803014103048601140804555897014469307122160867977
Probable prime factor 2 has 91 digits:
1120993897358460578404490502184135771743858065308635982721119473885677535372544659994092289

The linear algebra for 2,1044+ will finish tomorrow. Sieving for 2,1454L will
finish on Thursday. I will then do 2,1462L followed by 2,1478L.

I promised Sam Wagstaff that I would finish all the 2LM numbers through
768 bits. [roughly another 3 months]

What's the status of 3,479+? What will NFSNET do next?
The first few holes in the base 5 table have been around for a while.
Finishing base 3 to exponent 500 (one left; quite difficult) or working
on the first holes in the base 2 tables would be nice......

[xilman]

Originally Posted by R.D. Silverman

Finishing base 3 to exponent 500 (one left; quite difficult) or working on the first holes in the base 2 tables would be nice......

I've been vaguely assuming that the last base-3 number should be done next. At least, that was the purpose of discussion with Alex a few months back.

However, I'm not really in a position to contribute much sieving effort these days.


Paul