mersenneforum.org Team sieve #20: c170 from 4788:2549
 Register FAQ Search Today's Posts Mark Forums Read

 2010-11-05, 15:09 #144 EdH     "Ed Hall" Dec 2009 Adirondack Mtns 3,253 Posts Once this iteration is factored, is it a free-for-all to factor the next? Or, is there some coordinated effort to perform the (hopefully) trivial factoring of the next few composites? Would it be rude for those not post-processing to still run curves, or just time wasting. If a factor was happened upon, via the additional curves, would that be good or bad? I am not doing the above, but I'm restarting some machines after a trip and the questions came to mind.
2010-11-05, 16:26   #145
Mini-Geek
Account Deleted

"Tim Sorbera"
Aug 2006
San Antonio, TX USA

17×251 Posts

Quote:
 Originally Posted by EdH Once this iteration is factored, is it a free-for-all to factor the next? Or, is there some coordinated effort to perform the (hopefully) trivial factoring of the next few composites?
As far as I know, it's a free-for-all with the coordination being: factors are reported in the DB and significant work is reported in the thread. e.g. informing everyone how much ECM has been run on a number or saying that you will be completing a number via GNFS.
Quote:
 Originally Posted by EdH Would it be rude for those not post-processing to still run curves, or just time wasting. If a factor was happened upon, via the additional curves, would that be good or bad?
Do you mean, for example, to run ECM curves on this c170 right now? I'd call it wasteful, not rude. Either your ECM work will be wasted because no factor was found (which is most likely), or the GNFS work will be wasted because you found a factor (unlikely, but a net 'good' result).
They wouldn't be harmful, after all they just might find a factor and save us the extra couple days of post-processing, but the chances that you'd find one in this time period after the ECM we've given it is extremely slim, and whether you do the ECM or not, we'll have the answer very soon. If you were to find it via ECM now, then all the work done on the GNFS would have been wasted. But the end result is still that we have the factors.

Last fiddled with by Mini-Geek on 2010-11-05 at 16:34

 2010-11-05, 16:48 #146 em99010pepe     Sep 2004 283010 Posts You can run some ecm curves on this c170 but you have less than 71 hours until I get the factors from the post-processing.
 2010-11-05, 19:05 #147 Batalov     "Serge" Mar 2008 Phi(4,2^7658614+1)/2 29·313 Posts As shown by Tom's and Greg's tests, larger density filtering attempts will need 1. more relations to succeed (there's no free lunch) 2. more memory to run 3. but will be faster (in this size range) For every particular algebra setup, there's an optimum in the sky (probably the largest+densest matrix that still fits the memory and converges in filtering). The minimum total factoring time is even trickier - but doesn't have to be perfect if all participants have something else to switch to, anyway. I'd suggest having a few precompiled binaries at hand and try e.g. first, TD 100, and if it doesn't converge or the resulting matrix doesn't fit the memory, then fallback to 90 or 80. The good ol' 70 is also nothing to sneeze at. Sometimes it is simply the best (especially for aliquot run-of-the-mill transient gnfs projects). So, the algorithm: compile TD 70, 80, 90, 100, and also make of the not-LARGE_BLOCKS, just in case (because stages are separate, you can combine these two dimensions). Before every compilation, insert a self-descripting string in logprintf("Msieve 1.47..."), ok? We have this for example (B+D): Sun Oct 10 05:03:41 2010 Msieve v. 1.47 SVN379 LARGE_BLOCKS zlib density100 Sun Oct 10 05:03:41 2010 random seeds: 3abfbd9f cd113da7 Sun Oct 10 05:03:41 2010 factoring 2760869837544182879281843314...95882573313215568949924071 (211 digits) Sun Oct 10 05:03:42 2010 searching for 15-digit factors... etc My 2 cents. Last fiddled with by Batalov on 2010-11-05 at 19:07 Reason: /awfully long [and useless] string clipped/
 2010-11-05, 23:49 #148 jasonp Tribal Bullet     Oct 2004 3,527 Posts Looks like the filtering needs general command line argument parsing. Command lines for filtering and LA are in my queue. The target density defaults to 70 because it is a good compromise that keeps the memory use down while still producing a matrix that is small (enough) and sparse. My guess is that you have to be sieving with multiple machines, and generate a very large matrix, for a higher target density to reduce the LA time by more than the extra calendar-time spent sieving. Last fiddled with by jasonp on 2010-11-05 at 23:57
2010-11-06, 01:26   #149
Andi47

Oct 2004
Austria

2·17·73 Posts

Quote:
 Originally Posted by jasonp Looks like the filtering needs general command line argument parsing. Command lines for filtering and LA are in my queue. The target density defaults to 70 because it is a good compromise that keeps the memory use down while still producing a matrix that is small (enough) and sparse. My guess is that you have to be sieving with multiple machines, and generate a very large matrix, for a higher target density to reduce the LA time by more than the extra calendar-time spent sieving.
Hmmm... maybe keep a default of 70 (seems to be a good (enough) compromise, at least for small and medium size factorizations), but allow overriding this default by command line.

2010-11-06, 01:43   #150
Andi47

Oct 2004
Austria

2×17×73 Posts

Quote:
 Originally Posted by EdH Would it be rude for those not post-processing to still run curves, or just time wasting. If a factor was happened upon, via the additional curves, would that be good or bad?
just a few random thoughts about ECM during GNFS:

personally I am currently running sequence 10212 and currently I keep encountering cofactors around c140 to c145 (I know, that's quite a bit smaller than c170, but that's the difference between home computing and team computing.).

My approach is the following:

1.) ECM to the desired extent. (e.g. for a c141 I do full B1=11e6 and maybe 500@43e6)

2.) Poly search. In parallel (on the idle threads of my i7) I do P-1 and p+1 to e.g. B1=5e9 and B2=5e15, followed by yet more ECM, until poly search has finished.

3.) sieving. And only sieving. No ECM at this time. Not a single curve.

4.) postprocessing. At this time I might consider ECMing something else, but not the cofactor of the current cofactor of the sequence, hence I would consider it wasteful if I find no ECM factor, and very annoying if I would find an ECM factor just a few hours before a two-to-three-weeks-GNFS finishes to output the factors.

Last fiddled with by Andi47 on 2010-11-06 at 01:45

2010-11-06, 01:58   #151
Batalov

"Serge"
Mar 2008
Phi(4,2^7658614+1)/2

29·313 Posts

Quote:
 Originally Posted by EdH ...Would it be rude for those not post-processing to still run curves, or just time wasting. If a factor was happened upon, via the additional curves, would that be good or bad?
It won't be rude to run curves, -- but it would be rude to find a factor.

 2010-11-08, 11:19 #152 em99010pepe     Sep 2004 2×5×283 Posts LA will finish within 4 hours.
2010-11-08, 17:43   #153
em99010pepe

Sep 2004

2·5·283 Posts

Code:
prp55: 5020628089729196540791781957106211235328810013516737941
prp115: 6236805693823202812256436162373797339317645025210784211942726700957117792931768486741474427658616159852603436208839
ecm miss?
Attached Files
 msieve.zip (221.2 KB, 98 views)

Last fiddled with by em99010pepe on 2010-11-08 at 17:45

2010-11-09, 07:39   #154
schickel

"Frank <^>"
Dec 2004
CDP Janesville

2×1,061 Posts

Quote:
 Originally Posted by em99010pepe Code: prp55: 5020628089729196540791781957106211235328810013516737941 prp115: 6236805693823202812256436162373797339317645025210784211942726700957117792931768486741474427658616159852603436208839
Awesome job. Thanks for the assist with the post-processing.
Quote:
 Originally Posted by em99010pepe ecm miss?
I would say no. We had ~85% of t55 and it was getting hard to get any further.

I would have been bummed if it had turned out like my c146: three weeks to find out it was p43*p103.

Now that I've got it figured out, I can put an ECM server up on the next large composite. That'll make the job easier to track....

 Similar Threads Thread Thread Starter Forum Replies Last Post RichD Aliquot Sequences 29 2014-04-13 01:55 RichD Aliquot Sequences 49 2013-12-17 12:27 jrk Aliquot Sequences 14 2012-02-24 15:10 schickel Aliquot Sequences 51 2011-01-05 02:32 10metreh Aliquot Sequences 39 2009-05-03 14:02

All times are UTC. The time now is 21:20.

Mon Aug 3 21:20:29 UTC 2020 up 17 days, 17:07, 0 users, load averages: 1.84, 1.51, 1.41

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.