Fast Breeding (guru management) - Page 103

swellman · 2017-08-28, 14:07

Quote:

Originally Posted by VBCurtis

Swellman PM'ed me with a question about test-sieving C224_148_86, and I found it interesting so I took on test-sieving a few options:

Code:

n: 45627432792347670688239956412173506509008903107099345312947604301182520320088908379217379060309588664605324037213136495321294587557850433562888672922025539316571326365191223456037265379609488544611394330437749647736122392373
# 148^86+86^148, difficulty: 245.02, anorm: 2.55e+040, rnorm: 1.42e+046
# scaled difficulty: 245.02, suggest sieving rational side
# size = 8.583e-013, alpha = 0.000, combined = 1.120e-013, rroots = 0
type: snfs
size: 245
skew: 11.674
c6: 1
c0: 2531281
Y1: 4838313566980742053301602746368
Y0: -2303887567059225375711861260560690857602371813376
rlim: 268000000
alim: 268000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 64
rlambda: 2.8
alambda: 2.8

I tried 14e/32 and 14e/33 with the above parameters, also 15e/31 and 15e/32 with alim = rlim = 134M. All testing was done with Q-range of 2000 per test.
14e/32:

Code:

Q=30M   3237 rels   0.223 sec/rel
Q=80M   3136 rels   0.206 sec/rel
Q=130M  3290 rels   0.244 sec/rel
Q=180M  2626 rels   0.291 sec/rel
Q=230M  2766 rels   0.328 sec/rel
Q=280M  2772 rels   0.359 sec/rel
Target 450M relations => Q-range 20-330M suggested

14e/33:

Code:

Q=30M   5069 rels   0.142 sec/rel
Q=80M   5153 rels   0.126 sec/rel
Q=130M  5834 rels   0.138 sec/rel
Q=180M  4838 rels   0.158 sec/rel
Q=230M  5074 rels   0.179 sec/rel
Q=280M  5090 rels   0.197 sec/rel
Target 760M relations => Q-range 20-310M suggested

15e/31:

Code:

Q=30M   3968 rels   0.266 sec/rel
Q=80M   2869 rels   0.320 sec/rel
Q=130M  3024 rels   0.371 sec/rel
Target 240M relations => Q-range 20-170M suggested

15e/32:

Code:

Q=30M   6818 rels   0.155 sec/rel
Q=80M   5568 rels   0.165 sec/rel
Q=130M  5831 rels   0.192 sec/rel
Target 410M relations => Q-range 20-170M suggested

15e/32 is at least 35% faster than 14e/32 for the job (20-25% faster sec/rel at identical Q-values, but 15e avoids all the slow Q-range above 180M). I didn't do the integration to calculate estimated job length, but roughly 35% less time suggests 15e might be the right siever here.
We haven't tried 14e/33 yet, and 20-310M sieve range isn't a terrible idea; I estimate 14e/33 to be 10% faster job than 14e/32, with 15e/32 roughly 25% faster than 14e/33.

My thanks VBCurtis for the extensive trial sieving and analysis. Seems to me 15/31 is the way to go from your findings, but whatever folks want to try. A 14/33 sounds like masochism!

FWIW, the sievers were crashing with this number when using a new Win 10 machine. It's the only time I've seen this.

VBCurtis · 2017-08-28, 14:45

Quote:

Originally Posted by swellman

A 14/33 sounds like masochism!

FWIW, the sievers were crashing with this number when using a new Win 10 machine. It's the only time I've seen this.

Since 14/33 is untested ground, I'll do the LA if 14/33 is chosen by the gatekeepers.

My test-sieving was done in Ubuntu 14.04, so I can't comment on the Win-siever crash.

RichD · 2017-08-29, 12:48

QUEUED C165 from the OPN t550 file.
Sieve on the algebraic side.

Code:

n: 252200952570714959522276510841096370674323976995081889215727156126160616201246948514974489509681516961184893362356286486785968091610249508946327883536380191981616429
# 4635303278743^17-1, difficulty: 227.99, skewness: 129.13, alpha: 0.00
# cost: 1.47611e+18, est. time: 702.91 GHz days (not accurate yet!)
lss: 0
skew: 129.126
c6: 1
c0: -4635303278743
Y1: -1
Y0: 99594295370402687259450073008396038407
m: 99594295370402687259450073008396038407
rlim: 200000000
alim: 200000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7
type: snfs

Trial sieving 5K blocks.

Code:

swellman · 2017-08-30, 16:41

QUEUED C225_140_138 is ready for SNFS.

Code:

n: 542038016194898058377811481575996489349998488336457433551304722132283550121460835423485159142752243216163118385595486921487165241322448030181348476110253687710135764732748278610302647894409987865604739881721074443428491912449
# 140^138+138^140, difficulty: 258.07, anorm: 2.76e+038, rnorm: -3.26e+048
# scaled difficulty: 259.75, suggest sieving rational side
# size = 7.614e-013, alpha = 0.000, combined = 9.684e-014, rroots = 0
type: snfs
size: 258
skew: 5.1676
c6: 1
c0: 19044
Y1: -1965753632901132991452851230906979378964909
Y0: 2736874734008091634300000000000000000000000
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 on 14e.

Code:

Q=30M   9807
Q=80M   8073
Q=150M  7241
Q=250M  6208
Q=350M  5899

Suggesting a sieving range of 30M-340M

RichD · 2017-08-30, 21:17

QUEUED C189 from the OPN t550 file.
Sieve on the algebraic side.

Code:

n: 252594637447607245241185695796134629188854883886869844681604282726480736521809390738657412988155154855736991000412550165400686017997275949776088501372043867909702418929596732428615892729759
# 4052490063499^17-1, difficulty: 226.94, skewness: 126.27, alpha: 0.00
# cost: 1.35598e+18, est. time: 645.71 GHz days (not accurate yet!)
lss: 0
skew: 126.266
c6: 1
c0: -4052490063499
Y1: -1
Y0: 66552730150123655613864377003248440499
m: 66552730150123655613864377003248440499
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7
type: snfs

Trial sieving 5K blocks.

Code:

swellman · 2017-09-01, 01:02

C172_146_97 - GNFS - 14e/31 - posted in post 1120

C173_130_129 - GNFS - 15e/31 - posted in post 1122

C224_148_86 - SNFS - siever/lpb TBD - posted in post 1121

C224_148_86, well test sieved and analyzed by VBCurtis, is probably best sieved as a 15e/31 or 32 job, but as Curtis has suggested, a 14/33 is possible. Would this be a good opportunity to verify the speed/yield of such a job with a host of users willing to chew on such a non-trivial task? Probably not something to try everyday, but with the upcoming challenge the timing is perfect and the results could be instructive. Or has this topic been done to death in the past?

Of course if 14/33 is seen as abusing the goodwill of the sievers, then obviously the idea should be dropped.

VBCurtis · 2017-09-01, 01:18

There's a GNFS-179 in the 14e queue presently; that makes the GNFS173 from post 1122 a clear 14e choice.

I think it also makes 148_86 a clear 14e choice, given how much demand there is presently for workunits. I hope it's the first 14e/33!! :-)

fivemack · 2017-09-01, 08:26

I have queued 148_86 as a 14/33 job, and put VBCurtis down to do the post-processing. I don't have anything resembling an idea how big the post-processing will be.

Also queued the two GNFS jobs from #1128

I tend to do C17x jobs as 15/32 locally, but while we have a 14e queue it probably needs filling.

RichD · 2017-09-02, 02:09

QUEUED C185 from the OPN t550 file.
Sieve on the algebraic side.

Code:

n: 21937558583985708320116671473750602401633425908726074660018054001811997097813276941214646593868352842549839079152759241372197970005427726934544492415395022285746597747109414781668801143
# 1511140989109^17-1, difficulty: 219.23, skewness: 107.12, alpha: 0.00
# cost: 7.21829e+17, est. time: 343.73 GHz days (not accurate yet!)
lss: 0
skew: 107.123
c6: 1
c0: -1511140989109
Y1: -1
Y0: 3450761606696041645508981952839222029
m: 3450761606696041645508981952839222029
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7
type: snfs

Trial sieving 5K blocks.

Code:

fivemack · 2017-09-07, 16:44

I've added three of the smallest Fibonacci and Lucas SNFS numbers to the 15e queue.

The Fibonacci and Lucas GNFS numbers will take a little while longer because lots of them haven't had all that much ECM done yet, let alone the polynomial selection; also I am selfish and want to save some of them for my local cluster.

RichD · 2017-09-08, 13:35

QUEUED GNFS C157 from the t550 file.
This might be considered too small for 14e.

Code:

n: 1285513678558327974452479212785276193227825026334227699014956971970123507922903619704409921192405117718118925304817341669218102975406714504637552034617856703
# 113990869481^19-1 - GNFS(157)
lss: 0
Y0: -1176235284690866295875755403927
Y1: 1907379916459067
c0: -32000710470499238861821069593087061552
c1: -43772364993873314254620513196320
c2: 9509011773943995412695483
c3: -9946694495490369478
c4: 1843213838856
c5: 570960
skew: 3346989.46545
# lognorm 49.42, E 42.14, alpha -7.28 (proj -2.45), 3 real roots
# MurphyE=2.24660304e-12
rlim: 67000000
alim: 67000000
lpbr: 30
lpba: 30
mfbr: 60
mfba: 60
rlambda: 2.6
alambda: 2.6
type: gnfs

Trial sieving 5K blocks.

Code:

  Q  Yield
 20M 15996
 60M 15345
100M 13332

2017-09-01, 01:02	#1128
swellman Jun 2012 2²·773 Posts	Summarizing those waiting for enqueuing C172_146_97 - GNFS - 14e/31 - posted in post 1120 C173_130_129 - GNFS - 15e/31 - posted in post 1122 C224_148_86 - SNFS - siever/lpb TBD - posted in post 1121 C224_148_86, well test sieved and analyzed by VBCurtis, is probably best sieved as a 15e/31 or 32 job, but as Curtis has suggested, a 14/33 is possible. Would this be a good opportunity to verify the speed/yield of such a job with a host of users willing to chew on such a non-trivial task? Probably not something to try everyday, but with the upcoming challenge the timing is perfect and the results could be instructive. Or has this topic been done to death in the past? Of course if 14/33 is seen as abusing the goodwill of the sievers, then obviously the idea should be dropped. Last fiddled with by swellman on 2017-09-01 at 01:04

2017-09-01, 08:26	#1130
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 2³×11×73 Posts	I have queued 148_86 as a 14/33 job, and put VBCurtis down to do the post-processing. I don't have anything resembling an idea how big the post-processing will be. Also queued the two GNFS jobs from #1128 I tend to do C17x jobs as 15/32 locally, but while we have a 14e queue it probably needs filling. Last fiddled with by fivemack on 2017-09-01 at 08:26

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

2017-09-01, 01:18	#1129
VBCurtis "Curtis" Feb 2005 Riverside, CA 2×2,437 Posts	There's a GNFS-179 in the 14e queue presently; that makes the GNFS173 from post 1122 a clear 14e choice. I think it also makes 148_86 a clear 14e choice, given how much demand there is presently for workunits. I hope it's the first 14e/33!! :-)

2017-09-07, 16:44	#1132
fivemack (loop (#_fork)) Feb 2006 Cambridge, England 1100100011000₂ Posts	I've added three of the smallest Fibonacci and Lucas SNFS numbers to the 15e queue. The Fibonacci and Lucas GNFS numbers will take a little while longer because lots of them haven't had all that much ECM done yet, let alone the polynomial selection; also I am selfish and want to save some of them for my local cluster.