Planning & Coordination for 2,2330L - Page 9

vebis · 2019-05-06, 13:38

The other search finished with a better result:

Code:

n: 334377437706404684733884220732190564550147039308686365251932207592010868660163280944362994773042289837865752943086095333357861608005044927356022538834108845220872215597190097852981295940487497385774178731621
skew: 90705738.104
c0: -1655806548611327099132761402215592900502578257381
c1: -52470495453826760641954000116286955196049
c2: 2231551453970891751386654271420177
c3: 14227214675229453474894697
c4: -156074030035019784
c5: -209438460
Y0: -5445783017586151955857680467799412431984
Y1: 1986069033319259592352867
# MurphyE (Bf=3.436e+10,Bg=1.718e+10,area=5.469e+16) = 1.66e-08
# found by revision 9436c2ff8
# f(x) = -209438460*x^5-156074030035019784*x^4+14227214675229453474894697*x^3+2231551453970891751386654271420177*x^2-52470495453826760641954000116286955196049*x-1655806548611327099132761402215592900502578257381
# g(x) = 1986069033319259592352867*x-5445783017586151955857680467799412431984

There are also 2-3 decent polys in the archive.

VBCurtis · 2019-05-06, 14:06

Woohoo! Better, indeed:
skew 125540505.77018 score 1.64996072e-15

(from cownoise)

Have fun, Max!

Max0526 · 2019-05-07, 03:32

Quote:

Originally Posted by swellman

Best found so far by me, but not all the good...

Code:

# norm 2.488156e-20 alpha -7.890344 e 1.452e-15 rroots 5
skew: 1736805652.63
c0: -173423111471333766590048943026832487355179143421836
c1: 1415516976190808816842401925208948595004644
c2: 12769624767693438551061374008227397
c3: -34473379332757970694179620
c4: -4275141392861450
c5: 69300
Y0: -21712045787504165813577182256588814440415
Y1: 2692494731612253319

Maybe it can be spun in a good direction?

I can't make it better, sorry.

Max0526 · 2019-05-07, 03:35

@vebis
I saw both "a" and "b" archives today, I need some time to get through all of it, but I am on it.

Max0526 · 2019-05-07, 11:07

Quote:

Originally Posted by vebis

The other search finished with a better result:

Code:

n: 334377437706404684733884220732190564550147039308686365251932207592010868660163280944362994773042289837865752943086095333357861608005044927356022538834108845220872215597190097852981295940487497385774178731621
skew: 90705738.104
c0: -1655806548611327099132761402215592900502578257381
c1: -52470495453826760641954000116286955196049
c2: 2231551453970891751386654271420177
c3: 14227214675229453474894697
c4: -156074030035019784
c5: -209438460
Y0: -5445783017586151955857680467799412431984
Y1: 1986069033319259592352867
# MurphyE (Bf=3.436e+10,Bg=1.718e+10,area=5.469e+16) = 1.66e-08
# found by revision 9436c2ff8
# f(x) = -209438460*x^5-156074030035019784*x^4+14227214675229453474894697*x^3+2231551453970891751386654271420177*x^2-52470495453826760641954000116286955196049*x-1655806548611327099132761402215592900502578257381
# g(x) = 1986069033319259592352867*x-5445783017586151955857680467799412431984

There are also 2-3 decent polys in the archive.

After quick spin:

Code:

Y0: -5445783020173886700337776518838434019565
Y1: 7944276133277038369411468
c0: 24745062283834793771655911246007349221232259556
c1: 3638239206861172772594019092164033469860
c2: -543588643271079731014156451777817
c3: -15037081380967059802226545
c4: 618838392632323536
c5: 3351015360
skew: 32629082.89
# size 2.319e-20, alpha -7.604, combined = 1.671e-15 rroots = 5

Max0526 · 2019-05-07, 22:14

Another good poly from the "b" set after quick spin:

Code:

Y0: -6953261233697294683233295851584886174163
Y1: 5449957752208337629271462
c0: -1176061595709660314420928059407660349865025132200
c1: 37018662928190758651845113316638013040603
c2: 3268300215862651204474916762601600
c3: -38535057956097926521820731
c4: -316032307475148672
c5: 246874320
skew: 112240485.75
# size 2.361e-20, alpha -8.390, combined = 1.663e-15 rroots = 5

swellman · 2019-05-08, 00:22

Quote:

Originally Posted by R.D. Silverman

Perhaps the forum might want to do a poly search for the C210, C211, and C217?
If the forum provides polynomials perhaps Greg might queue them for lasievef?

Yoyo has queued the C210 2,2330M, all 72000 curves @B1=850M in fact. 2,2330L is on its final run to t65 as well. He is feeding the current BOINC competition with these tasks and a lot of others too.

The C211 and C217 are also on yoyo’s radar now, perhaps to be scheduled for ECM later this spring.

VBCurtis · 2019-05-09, 22:47

The machine that will host this job just finished sieving a C186 with CADO.
I=15, 32/33LP, Q=12-195M produced 675M raw relations, 450M unique.
About 6 hours of filtering (max memory use 31GB of 64GB physical) yielded a 22.0M matrix, which is now solving on CADO.
CADO's matrix step has 3 parts: krylov, lingen, mksol.
Krylov is using about 16GB, and will need 2 weeks to complete. lingen uses the most memory of the three, so I will wait for lingen to complete before considering firing up the CADO server for 2230L. Krylov is about 65% of the matrix step, so the job should yield factors by 1 June.

Back to 2230L:
I'm test-sieving the best-scoring deg 5 this weekend, while hoping Max's spin yields a 1.7 from Vebis' new batch. Hopefully, we can test client connections in 7-10 days, and start the project roughly 1 June.

Max0526 · 2019-05-09, 22:54

Quote:

Originally Posted by VBCurtis

...
Back to 2230L:
I'm test-sieving the best-scoring deg 5 this weekend, while hoping Max's spin yields a 1.7 from Vebis' new batch...

Doing it right now.

VBCurtis · 2019-05-11, 04:34

Quote:

Originally Posted by VBCurtis

Back to 2230L:
I'm test-sieving the best-scoring deg 5 this weekend, while hoping Max's spin yields a 1.7 from Vebis' new batch. Hopefully, we can test client connections in 7-10 days, and start the project roughly 1 June.

While I had a 4-5 day test-sieve already on the deg 6, the other tasks running on the machine have changed so an ETA-based speed measure of the deg 5 wouldn't be an accurate comparison.

So, I fired up simultaneous CADO instances, one 4-threaded sieve on deg 6 (a fresh run, not continuing the old one) and one on deg 5. Both at Q=200M. After about half a million relations found, deg 6 yields 5.6 or so, while deg 5 yields 6.1. The ETAs are amusingly similar: April 20 2039 for deg 6, April 3 2039 for deg 5. I guess we'll call it a tie.

I learned that reducing qmin on the snapshot file does not cause smaller Q-ranges to be added to the database. Then I tried changing qmin to 300M in a snapshot file; I'm pleased to report that the next workunit added to database was 300M! So, I'm comparing these two top-scoring polys at Q=300M overnight.
So, a multi-Q test sieve can be achieved by starting the project at the minimum Q. Run your test, note results, kill job. Edit snapshot file qmin to the next Q-value you wish to test, resume job. Yay!

VBCurtis · 2019-05-13, 03:23

Some more test-sieving shows the deg 6 poly is marginally faster at the same Q as the highest-scoring deg 5, but the deg 5 has 7-10% better yield. Better yield means more relations found at the lower & faster Q values, making up the 2-3% speed difference. Basically, it's a tie; we could choose the deg 6 if we wanted to do a 6th-degree job.

I tested I=15 vs I=16 at Q=300M. I=15 is some 30% faster! If we assume that we'll need 15% more relations from I=15 due to higher duplicate rate, there's still a strong argument for using I=15 for the higher Q values. In fact, I=15 is faster at Q=500M than I=16 is at Q=200M!

I currently plan on I=16 from Q=10M to 200M, and then I=15 for the remaining Q. This gives us higher yield on the most fruitful Q values, and faster sieving with less memory on the back half of the job. We should find ~1.3G relations from 10-200M at average yield of 6.5 or so, followed by 200-950M at average yield of just under 2.0 to get the other 1.4G relations. This combination (pending details like test-sieving 600 to 900M) appears to reduce expected sieve time to 45-48 core-years from a previous estimate of 52-55.

I'm testing Vebis' 1.65-score deg 5 posted last week presently.

2019-05-07, 03:35	#92
Max0526 "Max" Jun 2016 Toronto 1101111101₂ Posts	C207 polys @vebis I saw both "a" and "b" archives today, I need some time to get through all of it, but I am on it.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Coordination thread for redoing P-1 factoring	ixfd64	Lone Mersenne Hunters	81	2021-04-17 20:47
big job planning	henryzz	Cunningham Tables	16	2010-08-07 05:08
Sieving reservations and coordination	gd_barnes	No Prime Left Behind	2	2008-02-16 03:28
Sieved files/sieving coordination	gd_barnes	Conjectures 'R Us	32	2008-01-22 03:09
Special Project Planning	wblipp	ElevenSmooth	2	2004-02-19 05:25

2019-05-06, 14:06	#90
VBCurtis "Curtis" Feb 2005 Riverside, CA 2⁸×19 Posts	Woohoo! Better, indeed: skew 125540505.77018 score 1.64996072e-15 (from cownoise) Have fun, Max!

2019-05-09, 22:47	#96
VBCurtis "Curtis" Feb 2005 Riverside, CA 2⁸×19 Posts	The machine that will host this job just finished sieving a C186 with CADO. I=15, 32/33LP, Q=12-195M produced 675M raw relations, 450M unique. About 6 hours of filtering (max memory use 31GB of 64GB physical) yielded a 22.0M matrix, which is now solving on CADO. CADO's matrix step has 3 parts: krylov, lingen, mksol. Krylov is using about 16GB, and will need 2 weeks to complete. lingen uses the most memory of the three, so I will wait for lingen to complete before considering firing up the CADO server for 2230L. Krylov is about 65% of the matrix step, so the job should yield factors by 1 June. Back to 2230L: I'm test-sieving the best-scoring deg 5 this weekend, while hoping Max's spin yields a 1.7 from Vebis' new batch. Hopefully, we can test client connections in 7-10 days, and start the project roughly 1 June.

2019-05-13, 03:23	#99
VBCurtis "Curtis" Feb 2005 Riverside, CA 1300₁₆ Posts	Some more test-sieving shows the deg 6 poly is marginally faster at the same Q as the highest-scoring deg 5, but the deg 5 has 7-10% better yield. Better yield means more relations found at the lower & faster Q values, making up the 2-3% speed difference. Basically, it's a tie; we could choose the deg 6 if we wanted to do a 6th-degree job. I tested I=15 vs I=16 at Q=300M. I=15 is some 30% faster! If we assume that we'll need 15% more relations from I=15 due to higher duplicate rate, there's still a strong argument for using I=15 for the higher Q values. In fact, I=15 is faster at Q=500M than I=16 is at Q=200M! I currently plan on I=16 from Q=10M to 200M, and then I=15 for the remaining Q. This gives us higher yield on the most fruitful Q values, and faster sieving with less memory on the back half of the job. We should find ~1.3G relations from 10-200M at average yield of 6.5 or so, followed by 200-950M at average yield of just under 2.0 to get the other 1.4G relations. This combination (pending details like test-sieving 600 to 900M) appears to reduce expected sieve time to 45-48 core-years from a previous estimate of 52-55. I'm testing Vebis' 1.65-score deg 5 posted last week presently.