Current status

[bdodson]

Originally Posted by bdodson

Just noticed something I liked about Tom's new status page; unless
I'm imagining, it seems to have been updated already, with Dec 12
reln counts. ...

I've already noticed more that I like here. The daily CPU count went
from 93 on Sunday, to 90 then 89 then 86. I've taken the opportunity
to reset eight cpus that people had rebooted since I last did resets on
Sunday. I could probably pick up a few more, but it's finals week here;
prime-time for user complaints (these are Macs in our digital lab, running
the nfsnet version of the line siever), so I'll hold off for next week. One
can also see Saturday's cpu count down to 76; looks like there was an
extra quota of resets on Friday. Not that resets here are the only source
of variations; but my impression is that the other two large contributors
are more stable.

Again, here's your chance to make a bump in the cpu count; if our max
goes up above 93, that'd be due to your cpu(s). -Bruce

[em99010pepe]

Sam replied and as soon as he permits I'll post his reply...it's very funny. The link was added by him as a joke!!!!

Carlos

[R.D. Silverman]

Originally Posted by fivemack

NFSnet is presently working on 2^787-1, and has accumulated about 15 million relations, at a rate of a little over a million a day.

There is something bearing a vague resemblance to a status page at

http://www.chiark.greenend.org.uk/~twomack/nfsnet.html

but at present it is updated only manually; I have not yet (after nine months!) accumulated the effort to write the scripts to do automatic generation of the recent-contributions section.

I have finished sieving 6,299- and have started 2,776+.

[bdodson]

Originally Posted by fivemack

NFSnet is presently working on 2^787-1, and has accumulated about 15 million relations, at a rate of a little over a million a day.
...

Richard brings to my attention that Tom's stats are only tracking relations
from the line siever. Both Greg and now Richard are lattice sieving; and
the count yesterday was

... [we have] 50M unique relations and we are adding 4.5M per day ...

which is surely past halfway done. We're still welcoming people running
the line siever (without too much attention); as well as contributions from
experienced lattice sieving people. To recall, the number in question is
a base-2, M787, which has been the #1 most wanted number on Sam's
Cunningham page for some time. -Bruce

[Wacky]

I estimate about one more week for 2,787-.
During that time, I hope to get the version of the lattice siever that puts out tracking info from its current manually run state to something that will work like our linesiever.

[fivemack]

I've added that information to the stats page. Effectively distributed lattice sieving will be a fantastic addition to nfsnet; I suspect linear algebra will start to be the bottleneck on numbers not very much larger than the present more-wanteds.

I presume from the most recent Wagstaff circular that 2,787+ is the next project; what are the plans for the one after that?

[Wacky]

After 2.787+ ... well, that is next year :)

As for the LA being the bottleneck, that partly depends on Paul.
With at least 3 of us capable of handling a reasonably large matrix, I'm not sure that we cannot keep ahead of the sievers.

If we pick something sufficiently difficult, we might even be able to allow "off line" sievers to participate and still have time to send in their relations by banana boat.

6,383+ might be interesting ....

:)

[fivemack]

Originally Posted by Wacky

After 2.787+ ... well, that is next year :)

As for the LA being the bottleneck, that partly depends on Paul.
With at least 3 of us capable of handling a reasonably large matrix, I'm not sure that we cannot keep ahead of the sievers.

It all depends on the value of 'reasonably': the recent runs with matrix weights >500 million are no longer fitting quite as comfortably on 4G machines as earlier runs did, and >800 million looks to be getting troublesome, though 8G of 'slow' PC5300 RAM will fit on most core2 motherboards and costs $288 from crucial.

Fitting a curve to the last few runs suggests that weight hits 8e8 at about 825 bits and 1e9 around SNFS difficulty 255 (850 bits), which is just above anything in the current wanted list. I did some parameter selection for 2,841- and got an expected run-time with lattice sieving using the source code from ggnfs.sourceforge.net of about 12 GHz-years, or not very many weeks on a hundred modern CPUs.

6+383 is, I reckon, no more than 4 GHz-years; a minnow too small to be worthy of the appetite of the enhanced-NFSnet whale ... I hope

I am well into verifying that with proper choice of parameters a 516-bit GNFS is not significantly more than 1 GHz-year, and a large enough whale might be able to take a good bite into 170-digits territory. Sadly, 170-digits territory is fairly bare; 2^2154+1, given that 2,2154L is done and 2,2154M reduced to C171 ?

[xilman]

Originally Posted by fivemack

I am well into verifying that with proper choice of parameters a 516-bit GNFS is not significantly more than 1 GHz-year

I'm doing a 516-bit GNFS. The sieving is predicted to take ~110 days on a 2.2GHz AMD-64.

That is well under 1GHz-year. I doubt very much that the post-processing will take the total cost over 1 GHz-year.

Paul

[Wacky]

Well, we now have about 90M unique relations for 2,787-.
The sievers are switched to 2,787+. It should take a few weeks.

Greg reports the factors of
10,239- c228 as

prp54 factor: 383155477843726029783939406113226468701730728790004161
prp81 factor: 128780300340244872385688233345188210841783983757299260103530718169486826135819357
prp94 factor: 3290967632861131703281828943635774383301940171982919699073443165222894023742681701403432993547

[fivemack]

I've updated the nfsnet.html page; sorry, I'd got out of the habit over Christmas. Roughly how many relations have the lattice-sievers brought in to date?

How's the linear algebra for 2,787- going? Assuming there were no power outages over Christmas, it ought to be finished pretty soon.

Tom