mersenneforum.org  

Go Back   mersenneforum.org > Other Stuff > Archived Projects > ElevenSmooth

 
 
Thread Tools
Old 2008-04-08, 14:51   #45
bsquared
 
bsquared's Avatar
 
"Ben"
Feb 2007

2×1,789 Posts
Default

Quote:
Originally Posted by bsquared View Post
I'll take 90-92, with 15e.
done, and uploaded, with ~ 2.75M relations. I'll take 92-93. Any idea how much will be enough?
bsquared is offline  
Old 2008-04-10, 08:42   #46
jbristow
 
jbristow's Avatar
 
Aug 2007

3×31 Posts
Default

Paul's mentioned a rule of thumb of 0.8 * (pi(algebraic lp bound) + pi(rational lp bound)). For 30-bit large primes, that would come out at just over 87M relations (before duplicate removal, assuming no duplicated special q's). In my experience, msieve usually wants just a little bit more than this. It's been pretty stable across different composites for the same large prime bounds (although I haven't done a 30-bit lp factorization).

I just finished sieving a number with 29-bit large primes where this formula would predict 45.1M relations. 45.4M relations weren't enough, but it built at 46.5M. It sounds like you have a more sophisticated estimation process. How does it work?
jbristow is offline  
Old 2008-04-10, 10:49   #47
xilman
Bamboozled!
 
xilman's Avatar
 
"π’‰Ίπ’ŒŒπ’‡·π’†·π’€­"
May 2003
Down not across

10,949 Posts
Default

Quote:
Originally Posted by fivemack View Post
The models I have for duplicate removal (N_{duplicates removed} proportional to N_{input relations}^2) and for large ideals both said that I'd have lots of excess by this point and would be ready to start a matrix job, so I need better models.
That sounds very much like the birthday paradox estimator. I don't have a better model but it seems likely that the independent-uniform-random assumption underlying that estimator is not being met.

Palul
xilman is online now  
Old 2008-04-10, 13:10   #48
bsquared
 
bsquared's Avatar
 
"Ben"
Feb 2007

2·1,789 Posts
Default

Quote:
Originally Posted by jbristow View Post
Paul's mentioned a rule of thumb of 0.8 * (pi(algebraic lp bound) + pi(rational lp bound)). For 30-bit large primes, that would come out at just over 87M relations (before duplicate removal, assuming no duplicated special q's). In my experience, msieve usually wants just a little bit more than this. It's been pretty stable across different composites for the same large prime bounds (although I haven't done a 30-bit lp factorization).
Thanks for that rule of thumb, I hadn't seen that before.

Do the small prime bounds come into play at all, or is it akin to QS where as long as they aren't unreasonably small or obscenely huge they tend to not matter too much?
bsquared is offline  
Old 2008-04-10, 19:38   #49
xilman
Bamboozled!
 
xilman's Avatar
 
"π’‰Ίπ’ŒŒπ’‡·π’†·π’€­"
May 2003
Down not across

10,949 Posts
Default

Quote:
Originally Posted by bsquared View Post
Thanks for that rule of thumb, I hadn't seen that before.

Do the small prime bounds come into play at all, or is it akin to QS where as long as they aren't unreasonably small or obscenely huge they tend to not matter too much?
They don't matter at all subject to the constraints you mention.


Paul
xilman is online now  
Old 2008-04-10, 20:30   #50
jasonp
Tribal Bullet
 
jasonp's Avatar
 
Oct 2004

3·1,181 Posts
Default

Quote:
Originally Posted by bsquared View Post
Do the small prime bounds come into play at all, or is it akin to QS where as long as they aren't unreasonably small or obscenely huge they tend to not matter too much?
Bob's paper on SNFS provides more detail about this; the choice of factor base bound affects the relation discovery rate but it doesn't affect the number of relations you need, at least not nearly as much as the large prime bound.
jasonp is offline  
Old 2008-04-11, 08:11   #51
fivemack
(loop (#_fork))
 
fivemack's Avatar
 
Feb 2006
Cambridge, England

2·3·29·37 Posts
Default

I'm afraid something weird happened with the filtering using msieve-1.34: it went through many rounds of singleton removal, and then at the 'final stage of singleton removal' point it went from many ideals to 0 ideals - I suspect it's probably something simple like running out of disc space.

I don't think I have time to fix this before I leave the country, so: there are no more relations needed but I'll be happy with any that you send, I'll start the processing again on about 26 April, you might want to move to 3+512 where there is still a lot more sieving to do.
fivemack is offline  
Old 2008-04-11, 19:56   #52
fivemack
(loop (#_fork))
 
fivemack's Avatar
 
Feb 2006
Cambridge, England

643810 Posts
Default

It was just a matter of running out of disc space. I have restarted the job, it should finish while I'm in Ukraine and I should have results available before the end of April. I need no more relations.

Code:
found 24304315 hash collisions in 100541416 relations
found 19355979 duplicates and 81185437 unique relations
filtering rational ideals above 91947008
filtering algebraic ideals above 91947008
need 15968092 more relations than ideals
commencing singleton removal, pass 1
relations with 0 large ideals: 3755669
relations with 1 large ideals: 16664050
relations with 2 large ideals: 29440412
relations with 3 large ideals: 23833015
relations with 4 large ideals: 7486161
relations with 5 large ideals: 6130
relations with 6 large ideals: 0
relations with 7+ large ideals: 0
81185437 relations and about 57135683 large ideals

Last fiddled with by fivemack on 2008-04-11 at 20:50 Reason: log
fivemack is offline  
Old 2008-04-26, 19:25   #53
fivemack
(loop (#_fork))
 
fivemack's Avatar
 
Feb 2006
Cambridge, England

2×3×29×37 Posts
Default

I noticed, at 5:30am on Saturday two weeks ago just as I was about to get in the taxi to go to the airport, that the msieve -nc2 run had failed to start because there wasn't enough memory on the machine because it was also running four sievers for 3+512.

There wasn't enough time to fix this, so that machine has been sieving 3+512 for the last two weeks. Once the range I'm doing is finished, which will be another four days, I'll start a new msieve -nc2 run. Hope for a factor sometime around 10th May.
fivemack is offline  
Old 2008-04-28, 08:53   #54
fivemack
(loop (#_fork))
 
fivemack's Avatar
 
Feb 2006
Cambridge, England

643810 Posts
Default

Code:
Mon Apr 28 09:32:07 2008  matrix is 7933391 x 7933638 (2168.0 MB) with weight 536309417 (67.60/col)
Mon Apr 28 09:32:07 2008  sparse part has weight 488993426 (61.64/col)
Mon Apr 28 09:32:07 2008  matrix includes 64 packed rows
Mon Apr 28 09:32:07 2008  using block size 65536 for processor cache size 4096 kB
Mon Apr 28 09:33:00 2008  commencing Lanczos iteration (4 threads)
ETA is morning of Wed May 7th.
fivemack is offline  
 

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
M991 reservations thread fivemack Factoring 148 2015-06-19 03:34
Old reservations opyrt Prime Sierpinski Project 3 2009-03-26 19:51
SNFS parameters for M2376 fivemack ElevenSmooth 35 2008-03-28 20:06
Range reservations thread MooooMoo Twin Prime Search 38 2006-06-08 04:12
Deutscher Thread (german thread) TauCeti NFSNET Discussion 0 2003-12-11 22:12

All times are UTC. The time now is 07:41.


Sun Oct 17 07:41:23 UTC 2021 up 86 days, 2:10, 0 users, load averages: 1.16, 1.21, 1.23

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

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.