BOINC NFS sieving - NFS@Home - Page 91

Batalov · 2013-11-04, 18:25

I stand by my remark.

In it, "this difference" referred to the 5 or even 10-fold larger running time. See the original post. "Wow, what a difference, you go from 30bit to 31bit and the running time is 10 times longer!" This is not the case. Tom Womack ran some factorizations twice with the similar one-bit different LPBs and found minimal differences in running time (and matrix sizes). With a simple condition: both have to be comparably well-sieved (relative to the estimated minimum of relations which will be roughly 2x more, but the sieving can be 2x faster -- if all is done right and the conditions are not grossly contrived, like sieving with totally inappropriate LBPs).

Here's the real reason for 10-fold larger running time. One project was likely well-sieved and the other barely enough sieved (only enough for the proverbial "cusp of filtering convergence"). The 30-bit project may have also been much simpler, too; that would be another reason.

TL;DR version: if you would take a 30-bit project with SNFS-difficulty of 249, and then another project with SNFS-difficulty of 251 for which you would select 31-bit LPBs (or even the same project, once again*), _and_ sieve them both comparably, the wall clock time will not be very different (but apparently somewhat larger for the more difficult project. Somewhat larger, not 10 times larger).

_________
*but your name has to be Tom for that.

R.D. Silverman · 2013-11-04, 18:38

Quote:

Originally Posted by Batalov

I stand by my remark.

In it, "this difference" referred to the 5 or even 10-fold larger running time. See the original post. "Wow, what a difference, you go from 30bit to 31bit and the running time is 10 times longer!"

Ah. I missed that (i.e. the part about "10-fold"). I would normally expect a
matrix that is 20 to 25% larger in going from 30 to 31 bits.

Batalov · 2013-11-04, 18:49

Right. Tom could probably quantify that. (I haven't dug up his post, but it exists somewhere here on the forum. It was an interesting report. Could have been 6-7 years ago, now.)

VictordeHolland · 2013-11-04, 21:44

There are quite a few differences between this run and the two runs I did before:
SNFS(225) vs. GNFS(168)
30 bits vs. 31 bits
Relations: 110M+ vs. 220M+
Matrix size: ~1.5GB vs. 3.7GB

So I guess they all contribute to a longer LA phase?

Batalov · 2013-11-04, 23:02

Yes.

RichD · 2013-11-05, 16:25

I would like to reverse GW_4_369.

RichD · 2013-11-05, 19:35

Quote:

Originally Posted by RichD

I would like to reverse GW_4_369.

Oops, I am almost done downloading GC_4_369.
Please adjust my assignment.

RichD · 2013-11-06, 18:27

GC_4_369 splits as:

Code:

prp51 factor: 833216020977010133611098079725963902150797467457397
prp118 factor: 1540723616026788250128258003100454105963168284173198740572019759032558763687534559921149354430092646605553304218219221

wombatman · 2013-11-07, 19:44

I'll take a whack at C168_130_119.

swellman · 2013-11-07, 20:21

I'll grab it when it is ready for download.

Thanks.

wombatman · 2013-11-08, 05:03

ETA on the C168 is ~108 hours. And that's with the target density of the matrix at 100!

2013-11-06, 18:27	#998
RichD Sep 2008 Kansas D3F₁₆ Posts	GC_4_369 splits as: Code: prp51 factor: 833216020977010133611098079725963902150797467457397 prp118 factor: 1540723616026788250128258003100454105963168284173198740572019759032558763687534559921149354430092646605553304218219221

2013-11-07, 20:21	#1000
swellman Jun 2012 11×281 Posts	Reserving GC_5_318 I'll grab it when it is ready for download. Thanks.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Boinc Statistics for NFS@Home borked ?	thomasn	NFS@Home	1	2013-10-02 15:31
BOINC NFS sieving - RSALS	debrouxl	NFS@Home	621	2012-12-14 23:44
BOINC?	masser	Sierpinski/Riesel Base 5	1	2009-02-09 01:10
BOINC?	KEP	Twin Prime Search	212	2007-04-25 10:29
BOINC	bebarce	Software	3	2005-12-15 18:35

2013-11-04, 18:25	#991
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 9497₁₀ Posts	I stand by my remark. In it, "this difference" referred to the 5 or even 10-fold larger running time. See the original post. "Wow, what a difference, you go from 30bit to 31bit and the running time is 10 times longer!" This is not the case. Tom Womack ran some factorizations twice with the similar one-bit different LPBs and found minimal differences in running time (and matrix sizes). With a simple condition: both have to be comparably well-sieved (relative to the estimated minimum of relations which will be roughly 2x more, but the sieving can be 2x faster -- if all is done right and the conditions are not grossly contrived, like sieving with totally inappropriate LBPs). Here's the real reason for 10-fold larger running time. One project was likely well-sieved and the other barely enough sieved (only enough for the proverbial "cusp of filtering convergence"). The 30-bit project may have also been much simpler, too; that would be another reason. TL;DR version: if you would take a 30-bit project with SNFS-difficulty of 249, and then another project with SNFS-difficulty of 251 for which you would select 31-bit LPBs (or even the same project, once again), _and_ sieve them both comparably, the wall clock time will not be very different (but apparently somewhat larger for the more difficult project. Somewhat* larger, not 10 times larger). _________ *but your name has to be Tom for that.

2013-11-04, 18:49	#993
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 9,497 Posts	Right. Tom could probably quantify that. (I haven't dug up his post, but it exists somewhere here on the forum. It was an interesting report. Could have been 6-7 years ago, now.)

2013-11-04, 21:44	#994
VictordeHolland "Victor de Hollander" Aug 2011 the Netherlands 1176₁₀ Posts	There are quite a few differences between this run and the two runs I did before: SNFS(225) vs. GNFS(168) 30 bits vs. 31 bits Relations: 110M+ vs. 220M+ Matrix size: ~1.5GB vs. 3.7GB So I guess they all contribute to a longer LA phase?

2013-11-04, 23:02	#995
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 9,497 Posts	Yes.

2013-11-05, 16:25	#996
RichD Sep 2008 Kansas 6477₈ Posts	I would like to reverse GW_4_369.

2013-11-07, 19:44	#999
wombatman I moo ablest echo power! May 2013 1781₁₀ Posts	I'll take a whack at C168_130_119.

2013-11-08, 05:03	#1001
wombatman I moo ablest echo power! May 2013 6F5₁₆ Posts	ETA on the C168 is ~108 hours. And that's with the target density of the matrix at 100!