Fast Breeding (guru management) - Page 101

VBCurtis · 2017-08-14, 23:01

I agree that speed about doubles, but required relations only jump by 70%. That's why I've been submitting everything as 32LP jobs; it's 10%+ faster to get ~400M 32LP relations than it is to get ~240M 31LP relations, and those targets produce comparably-sized matrices for the SNFS jobs I've been submitting (roughly SNFS-240).

I found SNFS-232 to be equivalent in time between 31LP and 32LP, and I suspect 14e/33 is faster than 14e/32 for SNFS250+ (that's specifically what I'll be testing this fall). For GNFS, I think 165 or 166 is the size where 14e/32 is faster than 14e/31. Anyone interested in seeing raw relations vs matrix size for various 14e tasks might have a look at mklasson.com/factors, where I have reported relation counts and matrix sizes for a range of 14e (and some 15e) tasks. I built quite a few 32LP matrices (with TD 100+) with fewer than 300M raw relations!

swellman · 2017-08-15, 00:25

Quote:

Originally Posted by VBCurtis

I built quite a few 32LP matrices (with TD 100+) with fewer than 300M raw relations!

This is a key takeaway. I have always gone for almost double the relations on a 32 LP job and look for TD 130-140. Too gratuitous perhaps?

VBCurtis · 2017-08-15, 03:54

I think so, yes. My most-recent personal factorization was 13*2^824-1, an SNFS-250. I used 15e/32, and 342M raw relations built a 9.5M matrix at TD 124. 14e will have a higher duplicate rate, so 380M+ relations would have been required. 13_2_802m is currently in LA, with 385M raw 14e rels building a 9.2M matrix at TD 128.
I am unconvinced LA time is reduced much from TD's of 130+ vs 124, though 10% more relations often does build a notably smaller matrix. I just don't mind ~10M matrices, so I don't request 400M relations for my 32LP jobs. By the time I need 400M+ relations, I'll be trying 650M rels on a 33LP job.
My only 33LP data so far:
13_2_837m, SNFS-254, 15e 557M rels 13.3M matrix @TD 116
13_2_864m, SNFS-263, 15e 618M rels 15.8M matrix @TD 124
I think 14e will require 10% more raw relations, which is where I get 650M as a target for SNFS 250ish. I hope to build TD 130 from that!

unconnected · 2017-08-15, 11:22

C168 from aliquot sequence 11040:10047 is ready for GNFS:

Code:

# norm 3.283662e-16 alpha -7.189363 e 5.248e-13 rroots 3
n: 164736666196825435404787655557209541928453253817257190535489412880787204132992680104314895301014743595089954795439761491986708889605473014584152814138097043938952406623
lss: 0
skew: 109895634.01
c0: -2587514198810932133781061266797804601115248
c1: 118096329304515670253820808175119610
c2: -3227397132471174410574315469
c3: -3024371911208417764
c4: 537996052764
c5: 756
Y0: -737316031279832613957275011228637
Y1: 223134028521912307
rlim: 134217727
alim: 134217727
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7

Suggesting sieving range is 20-200M.

swellman · 2017-08-17, 11:08

C223_129_100 has completed a full t55 courtesy of yoyo@Home, plus more by me @B1=300M. It is ready for SNFS.

Code:

n: 4121964423333085982233645849696974555736659790176918393117155325578174843765145584449220313884341611314785419434214245850957549019875541011482943960868788283748164810375959133569837736384152647784409749435630112732651168981
# 129^100+100^129, difficulty: 260.10, anorm: 2.06e+039, rnorm: -1.59e+049
# scaled difficulty: 261.74, suggest sieving rational side
# size = 3.890e-013, alpha = 0.000, combined = 5.948e-014, rroots = 0
type: snfs
size: 260
skew: 10.1055
c6: 1
c0: 1065024
Y1: -5000000000000000000000000000000000000000000
Y0: 758621374683090977986568634824263809
rlim: 240000000
alim: 240000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 64
rlambda: 2.8
alambda: 2.8

Test sieving with Q in blocks of 5k

Code:

Q=20M    7495
Q=60M    6494
Q=120M   6081
Q=180M   4975
Q=250M   4732
Q=350M   4453

Suggesting a sieving range of 20M-390M.

VBCurtis · 2017-08-18, 00:59

13*2^809-1 has had ~0.65*t55 ECM, and is ready for the 14e queue:

Code:

#13*2^809-1 difficulty: 245
n: 1239789660577535690534614521677444030147477630236436423304522866975933818525373635094601468561768127340898032949877526163671525540683448874435752600175483996348771943183814887859855047267660475340958164569
m: 43556142965880123323311949751266331066368
type: snfs
skew: 0.74
c6: 13
c0: -2
rlim: 120000000
alim: 120000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 64
rlambda: 2.7
alambda: 2.7

Q=15M to 130M should produce ~400M relations:
Q=20M yield 4.5
Q=40M yield 4.0
Q=70M yield 3.3
Q=100M yield 3.2
Q=130M yield 2.9

I would like to do the LA for this.

swellman · 2017-08-18, 13:21

C223_134_130 is ready for SNFS using 14e.

Code:

n: 4417091095813892728688127286539418937001062060139144566410118517363873055548200137085311347969572389434795982006727158878482355239828084851904001712779423524660241818271804042718723366337991739188314563357310291511109941841
# 134^130+130^134, difficulty: 247.15, anorm: 6.45e+031, rnorm: 7.38e+054
# scaled difficulty: 250.99, suggest sieving rational side
# size = 6.455e-017, alpha = 0.000, combined = 1.666e-013, rroots = 1
type: snfs
size: 247
skew: 1.3236
c5: 16
c0: 65
Y1: -300569473249667159156751403864289752106501534569
Y0: 8885065410379260901802741051591932773590087890625
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7

Test sieving with Q in blocks of 5k

Code:

Q=20M    5224
Q=60M    5567
Q=100M   6099
Q=150M   5368
Q=200M   5313
Q=250M   4556

Suggesting a sieving range of 20M-240M

RichD · 2017-08-20, 21:17

C179 from OPN t550 file.
Sieve on the algebraic side.

Code:

n: 78661921744862268746392821910119971203310936316649614623464932083843089020842600468566329508985146010206665451770759873909581154091523093828626316045059995522213621312463847888383
# 3927085741^23-1, difficulty: 230.26, skewness: 39.72, alpha: 0.00
# cost: 1.77165e+18, est. time: 843.64 GHz days (not accurate yet!)
lss: 0
skew: 39.720
c6: 1
c0: -3927085741
Y1: -1
Y0: 237838158555059113223510634120175084561
m: 237838158555059113223510634120175084561
type: snfs
rlim: 132000000
alim: 132000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7
type: snfs

Trial sieving 5K blocks.

Code:

swellman · 2017-08-20, 22:18

C223_139_53 is ready for SNFS on 14e.

Code:

n: 2134306013540056436162362749944339187027634625746710385409150646699404123149356207283444594500512753512120063279898246384817114656023398427304457483745657908117396756331638727744896748005798167437682521600906082498653650293
# 139^53+53^139, difficulty: 241.40, anorm: 2.39e+034, rnorm: 2.91e+053
# scaled difficulty: 244.58, suggest sieving rational side
# size = 2.838e-017, alpha = 0.000, combined = 1.177e-013, rroots = 1
type: snfs
size: 241
skew: 42.7232
c5: 1
c0: 142337807
Y1: -2692452204196940400601
Y0: 1904251686165483303166544735375573842653419591761
rlim: 268000000
alim: 268000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 64
rlambda: 2.8
alambda: 2.8

Test sieving with Q in blocks of 5K

Code:

Q=30M    9733
Q=80M    9481
Q=150M   9547
Q=220M   9580
Q=300M   9436

Suggesting a sieving range of 30M-280M

Max0526 · 2017-08-21, 14:32

Quote:

Originally Posted by swellman

C173_129_85 is ready for GNFS, with many thanks to wombatman for the polynomial. I ended up going with a 14/31 job here, as 14/32 had yield >2.0 but took an estimated 53% longer to sieve (plus greatly increased postprocessing time and difficulty). 15/31 is likely optimal, but keeping the 14e queue fed seemed more important. Of course the mods can change it if they deem it necessary.

Code:

n: 16432623607473574219758563585400708524378818125206904056185988551457839250312821323878324356918801650702586763199236947949550568517635374514910417505008392250111636215513873
# norm 8.440626e-017 alpha -7.132528 e 2.296e-013 rroots 3
skew: 5870867.97
c0: 25419121017344360577864617013788268281760
c1: 28857530855456147761173668461688476
c2: 631397788691506677770489927
c3: -2664694777806407733134
c4: 223010337756792
c5: 14302440
Y0: -1028156805787591556952709321595503
Y1: 7449870672504553
rlim: 120000000
alim: 120000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7

Test sieving on the -a side, with Q in blocks of 10k

Code:

Q=20M    11058
Q=60M    10562
Q=100M   10791
Q=150M   10918
Q=200M   9900
Q=250M   8316

Suggesting a sieving range of 20M-280M

I am back. I ran wombatman's polys through CADO and got 2 new ones with higher E:

Code:

Y0: -1009427825229151996551657342942778
Y1: 210906019054079
c0: 1517916384314284524046593239944919275305
c1: 51606612902664333569121720673602967
c2: -21530506587449528039635915175
c3: -3400396043158328006231
c4: 285057466029006
c5: 15679440
skew: 7472683.89357
# lognorm 55.38, E 47.69, alpha -7.69 (proj -2.60), 5 real roots
# MurphyE=2.41438948e-13

Code:

Y0: -3197822479513387219745948715313154
Y1: 1754259244774553
c0: 21923856716069276138650845587215150125578775
c1: 11093238189155958142431413095849715939
c2: -139660187299407105214765983455
c3: -2538515809629993061807
c4: 2171373063408
c5: 49140
skew: 122161020.18715
# lognorm 56.45, E 48.25, alpha -8.20 (proj -2.21), 5 real roots
# MurphyE=2.33756938e-13

Hope it's not too late.

Max0526 · 2017-08-21, 15:43

Code:

R0: -1009427825229211597749017949451862
R1: 210906019054079
A0: -14706812234442561584746077625649079501315
A1: 62935577997654071065756497728660415
A2: -18514641213973620944626070771
A3: -3710098762779786734135
A4: 262902730897806
A5: 15679440
skew 7500107.12, size 7.565e-017, alpha -7.687, combined = 2.415e-013 rroots = 5

Happy test-sieving!

2017-08-15, 03:54	#1103
VBCurtis "Curtis" Feb 2005 Riverside, CA 2·2,437 Posts	I think so, yes. My most-recent personal factorization was 132^824-1, an SNFS-250. I used 15e/32, and 342M raw relations built a 9.5M matrix at TD 124. 14e will have a higher duplicate rate, so 380M+ relations would have been required. 13_2_802m is currently in LA, with 385M raw 14e rels building a 9.2M matrix at TD 128. I am unconvinced LA time is reduced much from TD's of 130+ vs 124, though 10% more relations often does build a notably smaller matrix. I just don't mind ~10M matrices, so I don't request 400M relations for my 32LP jobs. By the time I need 400M+ relations, I'll be trying 650M rels on a 33LP job. My only 33LP data so far: 13_2_837m, SNFS-254, 15e 557M rels 13.3M matrix @TD 116 13_2_864m, SNFS-263, 15e 618M rels 15.8M matrix @TD 124 I think 14e will require 10% more raw relations, which is where I get 650M as a target for SNFS 250ish. I hope to build TD 130 from that! Last fiddled with by VBCurtis on 2017-08-15 at 04:08 Reason: added _2_ for clarity*

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
System management notes	kriesel	kriesel	7	2020-10-21 18:52
Improving the queue management.	debrouxl	NFS@Home	10	2018-05-06 21:05
Script-based Primenet assignment management	ewmayer	Software	3	2017-05-25 04:02
Do normal adults give themselves an allowance? (...to fast or not to fast - there is no question!)	jasong	jasong	35	2016-12-11 00:57
Power Management settings	PrimeCroat	Hardware	3	2004-02-17 19:11

2017-08-14, 23:01	#1101
VBCurtis "Curtis" Feb 2005 Riverside, CA 2×2,437 Posts	I agree that speed about doubles, but required relations only jump by 70%. That's why I've been submitting everything as 32LP jobs; it's 10%+ faster to get ~400M 32LP relations than it is to get ~240M 31LP relations, and those targets produce comparably-sized matrices for the SNFS jobs I've been submitting (roughly SNFS-240). I found SNFS-232 to be equivalent in time between 31LP and 32LP, and I suspect 14e/33 is faster than 14e/32 for SNFS250+ (that's specifically what I'll be testing this fall). For GNFS, I think 165 or 166 is the size where 14e/32 is faster than 14e/31. Anyone interested in seeing raw relations vs matrix size for various 14e tasks might have a look at mklasson.com/factors, where I have reported relation counts and matrix sizes for a range of 14e (and some 15e) tasks. I built quite a few 32LP matrices (with TD 100+) with fewer than 300M raw relations!