mersenneforum.org  

Go Back   mersenneforum.org > Factoring Projects > NFS@Home

Reply
 
Thread Tools
Old 2022-09-10, 06:03   #122
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

22×3×457 Posts
Default

Behavior should change at higher Q; a single test-point may mislead you, especially one chosen that makes GGNFS change alim down to 20M. Or are you sieving the r-side with 3LP on the a side?
I suppose I'm a bit surprised 3LP is useful at all at this size, too. Seems I have some testing to do!
VBCurtis is offline   Reply With Quote
Old 2022-09-10, 14:41   #123
RichD
 
RichD's Avatar
 
Sep 2008
Kansas

368210 Posts
Default

Parameters updated and Q range lowered to 18-96M

C234 from the OPN t800 file.
Increasing alambda improved the yield and cut the time.
3.5 & 3.6 were pretty close until I got to higher Q, then I gained a bit of time.
Code:
n: 10512896323946525328323590051713030182707488022156443686427827035707600086867033903192945127273531284579440671455984668440538086981016196248405396732898676231093060782972818311914352957505254809179243
# 34210243^31-1, difficulty: 233.56, skewness: 0.06, alpha: 0.00
# cost: 2.30621e+18, est. time: 1098.20 GHz days (not accurate yet!)
skew: 18.02
c6: 1
c0: -34210243
Y1: 46857681410903935241309012906079659443
Y0: -1
type: snfs
rlim: 67000000
alim: 67000000
lpbr: 30
lpba: 30
mfbr: 60
mfba: 87
rlambda: 2.6
alambda: 3.5
Trial sieving 2K blocks.
Code:
  Q  Yield N-Yld
 18M  4531  4237
 20M  4534  4386
 50M  3207  3381
 80M  2432  2813
 96M  2874  2696
Estimated total yield is about 130M relations from 18-96M.

Last fiddled with by swellman on 2022-09-11 at 02:49
RichD is offline   Reply With Quote
Old 2022-09-10, 16:00   #124
chris2be8
 
chris2be8's Avatar
 
Sep 2009

3×797 Posts
Default

Quote:
Originally Posted by RichD View Post
No problem. lambda is the least understood parameter in my knowledge suite.
Most of what I know about lambda is from reading the INSTALL.and.USE (or cweb.INSTALL.and.USE) file shipped with the lattice siever. Read the paragraph after the example poly a few times and it should start to make sense. (The first sentence is "The selection of the third field (called lambda) of the last two lines requires some care.")

I found setting alambda to mfba/lpba+0.6 and rlambda to mfbr/lpbr+0.6 is usually close to optimum. Test sieve about 0.1 higher and lower if you want to be sure.
chris2be8 is offline   Reply With Quote
Old 2022-09-10, 18:45   #125
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

10101011011002 Posts
Default

We don't need to guess, really; Charybdis has provided the formula:
mfb * log2 / log(lim), for each side. Add 0.2 (maybe 0.15) to this formula to get lambda. Note that since the logs are a ratio, it doesn't matter what base of log you use. If you choose log base 2, the formula simplifies to mfb / log_2 (lim).

We use the same mfb's and lim's over and over, so there should indeed be a "normally best" choice for lambda. But rather than one a single guess for test-sieving, we can use this formula to calculate what should be best, test it, and gain new best practices and new efficiencies.

The bigger the job, the more thorough we should be about test-sieving to select fastest parameters. Particularly on the f-small queue, it's a scarce resource so we should take care to maximize efficiency to get more jobs through the queue.

Finally, GGNFS does not sieve below the factor base bound. So, when Q is below lim on the sieving side, lim is reduced to Q+1 for that current Q-range. This means that lambda on the sieve side can be profitably set a bit higher than theory might indicate, because for small Q the lim is actually smaller than the one we plug into the above formula. So, if you test-sieve only at starting Q, large lambdas look faster (this is how some regulars came to use 3.0 for lambda on the sieve side for big jobs!). Choosing a slightly smaller lambda on the sieve side will miss a few relations at small Q, but increase speed quite a lot at higher Q for a net gain in job efficiency.

Last fiddled with by VBCurtis on 2022-09-10 at 18:46
VBCurtis is offline   Reply With Quote
Old 2022-09-18, 17:31   #126
swellman
 
swellman's Avatar
 
Jun 2012

368710 Posts
Default

QUEUED AS 8m7_293

8-7,293 is a SNFS 264 composite from the HCN project now ready for sieving on 15e_small.

Code:
n: 4176064904988291668319256684245334520819225751970049374156458143676426538538662977551445590104482819335306583796700863207451122460633606955455840311860352935559349870845679008170278701780721552630915396231909360854619140673310487581055982647862666927
skew: 1.0225
type: snfs
size: 264
c6: 7
c0: -8
Y1: -256923577521058878088611477224235621321607
Y0: 178405961588244985132285746181186892047843328
rlim: 134000000
alim: 134000000
lpbr: 32
lpba: 32
mfbr: 94
mfba: 64
rlambda: 3.5
alambda: 2.4
Results of test sieving on the rational side with Q in blocks of 1000:

Code:
MQ        Norm_yield      Speed (sec/rel)
30           3030              0.228
35           3286              0.229
50           3086              0.240
75           2732              0.212
100          2435              0.242
150          2232              0.305
200          1962              0.326
Suggesting a sieving range for Q of 30-205M to generate 435M raw relations.

Last fiddled with by swellman on 2022-09-18 at 17:42
swellman is online now   Reply With Quote
Old 2022-09-19, 01:58   #127
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

125548 Posts
Default

QUEUED AS 5_2_797m1

5*2^797-1 is ready for e-small after a bit more than t50 ECM:
Code:
n: 270433155750158790242010892459501546427209685877511991891679137291088366968633450805453957734005590934784516146721521454135745940896577442627145353523255306467821546083685971277
type: snfs
skew: 0.87
c6: 5
c0: -2
Y1: -1
Y0: 10889035741470030830827987437816582766592
rlim: 67000000
alim: 50000000
lpbr: 32
lpba: 31
mfbr: 60
mfba: 59
rlambda: 2.5
alambda: 2.7
Test-sieving 1kQ blocks on the -a side (r side was 5% worse sec/rel and yield) suggests Q from 10-72M should be enough for 235M rels:
Code:
#Q=10M	4525 (0.056 sec/rel) 60 spq
#Q=20M	4723 (0.060 sec/rel) 64 spq
#Q=30M	4564 (0.069 sec/rel) 69 spq
#Q=40M	2658 (0.074 sec/rel) 42 spq
#Q=50M	3544 (0.083 sec/rel) 60 spq
#Q=60M	3761 (0.083 sec/rel) 64 spq
#Q=70M	2805 (0.095 sec/rel) 54 spq
I'd like to handle the matrix.

Last fiddled with by swellman on 2022-09-19 at 11:08
VBCurtis is offline   Reply With Quote
Old 2022-09-19, 18:10   #128
swellman
 
swellman's Avatar
 
Jun 2012

3·1,229 Posts
Default

QUEUED AS 8p7_293

8+7,293 is a SNFS 264 composite from the HCN project now ready for sieving on 15e_small.

Code:
n: 1967645418902556993102340205275509177859691737980577693407882376486262548443738138144152558845475959819652761346438786973419287261202756248682338401729752076187830692977934455064562990990335070041661615644737980618799654264668976711817488334888088498029513
skew: 1.0225
type: snfs
size: 264
c6: 7
c0: 8
Y1: -256923577521058878088611477224235621321607
Y0: 178405961588244985132285746181186892047843328
rlim: 134000000
alim: 134000000
lpbr: 32
lpba: 32
mfbr: 94
mfba: 64
rlambda: 3.5
alambda: 2.4
Results of test sieving on the rational side with Q in blocks of 1000:

Code:
MQ       Norm_yield
35          3755
50          3571
75          3229
100         2807
150         2615
200         2242
Suggesting a sieving range for Q of 35-190M to generate 450M raw relations.

Last fiddled with by swellman on 2022-09-19 at 19:17
swellman is online now   Reply With Quote
Old 2022-09-22, 23:06   #129
unconnected
 
unconnected's Avatar
 
May 2009
Moscow, Russia

3·941 Posts
Default

QUEUED AS C168_159978_11480

C168 from 159978:i11480 for 15e_small queue:
Code:
n: 479997675371566508012116397954554151185920157133785715995605902715734356071413419630755029113057569989638996427419445937752909467028151487665362583627381236587716882241
# norm 2.561883e-16 alpha -8.506202 e 4.441e-13 rroots 5
skew: 105355368.15
c0: 3244787399662329815032658475396031643546363
c1: 1938884008252625169236175878029579329
c2: -68628897256567631078463320677
c3: -521359176148822777161
c4: 5409824979674
c5: 12792
Y0: -518632982182829025328913494470354
Y1: 65478461370689297
rlim: 67000000
alim: 67000000
lpbr: 30
lpba: 30
mfbr: 60
mfba: 60
rlambda: 2.6
alambda: 2.6
type: gnfs
lss: 0
Suggesting sieving range is 16M-80M, I'll take the LA.

Last fiddled with by swellman on 2022-09-23 at 00:03
unconnected is online now   Reply With Quote
Old 2022-09-23, 10:35   #130
RichD
 
RichD's Avatar
 
Sep 2008
Kansas

2×7×263 Posts
Default

QUEUED AS 62478139_29m1

C205 from the OPN t800 file.
Code:
n: 4487849528849322742479423745168653711324738985774913403224381060832884261299856002696991515300078883810194269384816276032531675031611350339460955093377981887727976150814972271321220207483134545019447746679
# 62478139^29-1, difficulty: 233.87, skewness: 19.92, alpha: 0.00
# cost: 2.36433e+18, est. time: 1125.87 GHz days (not accurate yet!)
skew: 19.920
c6: 1
c0: -62478139
Y1: -1
Y0: 952007620815896289237318398880046834699
type: snfs
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 90
rlambda: 2.7
alambda: 3.4
Trial sieving 2K blocks.
Code:
  Q  Yield N-Yld
 20M  8004  7742
 60M  5877  6133
100M  5238  5170

Last fiddled with by swellman on 2022-09-23 at 13:34
RichD is offline   Reply With Quote
Old 2022-09-28, 22:32   #131
RichD
 
RichD's Avatar
 
Sep 2008
Kansas

2·7·263 Posts
Default

QUEUED AS 177__227_5m1

QUEUED AS 177__227_5m1_2 on 15e_small

C193 from the OPN t800 file.
Code:
n: 4274594826568614723593930959135796200110466876118853098160362588328709964195214156563279817414634091541337984597138174381423495330370822349670629349298241613601983274066286183455653260622286221
# 17711138627434147625517360826415227976641556334393747227^5-1, difficulty: 220.99, skewness: 1.00, alpha: 1.45
# cost: 8.34884e+17, est. time: 397.56 GHz days (not accurate yet!)
skew: 1.000
c4: 1
c3: 1
c2: 1
c1: 1
c0: 1
Y1: -1
Y0: 17711138627434147625517360826415227976641556334393747227
type: snfs
rlim: 67000000
alim: 67000000
lpbr: 31
lpba: 31
mfbr: 90
mfba: 62
rlambda: 3.5
alambda: 2.7
Trial sieving 2K blocks.
Code:
  Q  Yield N-Yld
 20M  6780  6558
 60M  8695  9074
 80M  7327  8475
 96M  9216  8645

Last fiddled with by swellman on 2022-09-29 at 00:10
RichD is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
2022 Queue management of 15e swellman NFS@Home 129 2022-09-30 17:00
Queue management for 14e queue VBCurtis NFS@Home 140 2022-09-20 17:33
Queue management for 16e queue VBCurtis NFS@Home 147 2022-09-08 23:54
Queue management for e_small and 15e queues VBCurtis NFS@Home 254 2022-01-02 01:59
Improving the queue management. debrouxl NFS@Home 10 2018-05-06 21:05

All times are UTC. The time now is 11:01.


Mon Oct 3 11:01:38 UTC 2022 up 46 days, 8:30, 0 users, load averages: 1.58, 1.18, 1.11

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, 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.

≠ ± ∓ ÷ × · − √ ‰ ⊗ ⊕ ⊖ ⊘ ⊙ ≤ ≥ ≦ ≧ ≨ ≩ ≺ ≻ ≼ ≽ ⊏ ⊐ ⊑ ⊒ ² ³ °
∠ ∟ ° ≅ ~ ‖ ⟂ ⫛
≡ ≜ ≈ ∝ ∞ ≪ ≫ ⌊⌋ ⌈⌉ ∘ ∏ ∐ ∑ ∧ ∨ ∩ ∪ ⨀ ⊕ ⊗ 𝖕 𝖖 𝖗 ⊲ ⊳
∅ ∖ ∁ ↦ ↣ ∩ ∪ ⊆ ⊂ ⊄ ⊊ ⊇ ⊃ ⊅ ⊋ ⊖ ∈ ∉ ∋ ∌ ℕ ℤ ℚ ℝ ℂ ℵ ℶ ℷ ℸ 𝓟
¬ ∨ ∧ ⊕ → ← ⇒ ⇐ ⇔ ∀ ∃ ∄ ∴ ∵ ⊤ ⊥ ⊢ ⊨ ⫤ ⊣ … ⋯ ⋮ ⋰ ⋱
∫ ∬ ∭ ∮ ∯ ∰ ∇ ∆ δ ∂ ℱ ℒ ℓ
𝛢𝛼 𝛣𝛽 𝛤𝛾 𝛥𝛿 𝛦𝜀𝜖 𝛧𝜁 𝛨𝜂 𝛩𝜃𝜗 𝛪𝜄 𝛫𝜅 𝛬𝜆 𝛭𝜇 𝛮𝜈 𝛯𝜉 𝛰𝜊 𝛱𝜋 𝛲𝜌 𝛴𝜎𝜍 𝛵𝜏 𝛶𝜐 𝛷𝜙𝜑 𝛸𝜒 𝛹𝜓 𝛺𝜔