Fast Breeding (guru management) - Page 117

wombatman · 2017-11-30, 23:44

Quote:

Originally Posted by VBCurtis

There's no doubt in my mind that you want 33LP over 32; I'm pretty confident 34LP would be faster, and I would test 35 if I were running this factorization myself. LP bounds above 33 require non-standard sievers, either 16f, or the special 16e compilation floating around the forum that has the 33-bit LP bound removed.

16e is limited by 96 for mfbr/a in any case, so 3 large primes is limited to 33/96 on any 16e siever. For the 2LP side, 34/67 and 34/68 would be interesting to test; maybe I'll try that on your composite tonight, as I have some free time.

Sounds good. I'll be interested to see how that goes.

swellman · 2017-12-01, 13:41

QUEUED C246_143_58 is ready for SNFS on 15e.

Code:

n: 202234776336417710261405650689553937354003752024559559819720848415479775339953852271735016014830768330201249099498810443191526804931744370369043836242022676379833111285928285371688174040550889129206034422099044852193382687944629320767783463076199
# 143^58+58^143, difficulty: 257.46, anorm: 1.51e+036, rnorm: 1.81e+055
# scaled difficulty: 260.64, suggest sieving rational side
# size = 1.911e-018, alpha = 0.000, combined = 2.190e-014, rroots = 1
type: snfs
size: 257
skew: 1.7185
c5: 195112
c0: 2924207
Y1: -511324276025564512546607
Y0: 23767517358231570773047645414309870043308402671616
rlim: 268000000
alim: 268000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 64
rlambda: 2.8
alambda: 2.8

Test sieving on the -r side with Q in blocks of 2K.

Code:

30M    2908
80M    2678
150M   3017
250M   3066
350M   2801

Suggesting a sieving range for Q of 30M-350M with target # rels=460M.

unconnected · 2017-12-01, 17:39

QUEUED AS C175_11040_10071 C175 from 11040:10071 (thx for excellent poly was found by VBCurtis)

Code:

n: 6484689970303129020517057103365894793216912303102493240993925500641398178923457820535718262847740804273820043327446358642419268046237801717129218758317768624690555974293740733
# norm 6.007105e-17 alpha -6.096724 e 1.935e-13 rroots 1
skew: 23869212.60
c0: 215515253399468621655935634084010628842875
c1: 18221026257270678573040535301222710
c2: 810840940444709291461562333
c3: -92370495278835169378
c4: -2376612213148
c5: 147408
Y0: -8485425654983521960650346075265192
Y1: 20565115814604599
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7
type: gnfs

Suggesting sieving range 20M-140M with 15e siever.
I'll do LA for it.

LaurV · 2017-12-02, 01:59

Quote:

Originally Posted by VBCurtis

This!
<snip>
Hope this helps!

It helped me too, to understand few things that were unclear for me.

wombatman · 2017-12-02, 03:38

Reposting the final parameters and timings for the C208 blocker:

Code:

n: 8095101662371927421703337019465587498085337648622133688278589711654019359923503887978141510461468343349838217540569173400647791769725685803537804186347867144149599002247585690859122186539724272741806859085719
skew: 771127364.56
Y0: -17068243492239505219994785346910834818341
Y1: 1873940548553722757
c0: 165792391853474935561243616954647727516748946250496
c1: 2160239644350504494844955872920952825447896
c2: -21514458180493538566295548810659238
c3: -5887571126475837688637761
c4: 35919796435243602
c5: 5588280
type: gnfs

rlim: 800000000
alim: 800000000
lpbr: 33
lpba: 33
mfbr: 65
mfba: 96
rlambda: 2.6
alambda: 3.6

Final Q-block timings:

 50M: total yield: 1888, q=50002009 (2.61830 sec/rel)
100M: total yield: 3060, q=100002011 (2.20105 sec/rel)
200M: total yield: 2736, q=200002007 (2.31068 sec/rel)
300M: total yield: 3433, q=300002029 (2.45759 sec/rel)
400M: total yield: 2929, q=400002011 (2.65905 sec/rel)
500M: total yield: 2690, q=500002003 (2.87627 sec/rel)
600M: total yield: 2548, q=600002003 (2.86426 sec/rel)
700M: total yield: 2231, q=700002011 (3.56041 sec/rel)
800M: total yield: 2877, q=800002003 (3.12972 sec/rel)

As noted previously, the C207 used ~950M relations. I estimated a q-range of about 635M being needed before, and I think that holds decently well. So recommending sieving the q=100M-800M range at the least, and one could search a bit lower and a bit higher as well, as 900M sieved decently before.

VBCurtis · 2017-12-02, 22:48

Quote:

Originally Posted by wombatman

Sounds good. I'll be interested to see how that goes.

Well, I tried 7 different parameter settings before realizing that sieving the rational side is the less-efficient side. I'm repeating some of the choices on the -a side now; I should have data tomorrow.
Data for the best 3 choices:
65/96 -r 2067 rels, 3.67 sec/rel (33LP)
67/96 -r 4025 rels, 1.87 sec/rel (34LP) #looks best of -r options
69/96 -r 4970 rels, 1.66 sec/rel (35/34)

VBCurtis · 2017-12-03, 05:30

Data on the -a sieving side for the C208:
69/96 -a: 7134 rels, 1.05 sec/rel (35/34)
67/96 -a: 5418 rels, 1.43 sec/rel (34LP)
65-96 -a: 2736 rels, 2.77 sec/rel (33LP)

going to 34LP on both sides, even though mfbr is capped at 96, doubles yield and nearly cuts sec/rel in half. Since only 70% more relations are needed, this would result in project length 1.70/1.94 (ratio of extra rels needed to sec/rel ratio), which is around 86%. So, going to 34LP would save 14% of total sieving time. I would aim for 1600-1700M raw relations.

If Frmky were interested in running as 34, a couple other Q values should be tested to make sure 34 is always faster.

Going to 35LP on the r side should require 30% more relations, 2100-2200M raw relations. Yield on this test is over 30% higher, and sec/rel is improved by more than 30% also. 35/34 happens to be what CADO would choose for a C210 (they have default settings for every 5 digits, 208 is closest to 210). I'm not convinced this is improved enough to bother trying 35, which is a stretch, but I'd do 34 on both sides for sure!

wombatman · 2017-12-03, 06:02

Quote:

Originally Posted by VBCurtis

Data on the -a sieving side for the C208:
69/96 -a: 7134 rels, 1.05 sec/rel (35/34)
67/96 -a: 5418 rels, 1.43 sec/rel (34LP)
65-96 -a: 2736 rels, 2.77 sec/rel (33LP)

going to 34LP on both sides, even though mfbr is capped at 96, doubles yield and nearly cuts sec/rel in half. Since only 70% more relations are needed, this would result in project length 1.70/1.94 (ratio of extra rels needed to sec/rel ratio), which is around 86%. So, going to 34LP would save 14% of total sieving time. I would aim for 1600-1700M raw relations.

If Frmky were interested in running as 34, a couple other Q values should be tested to make sure 34 is always faster.

Going to 35LP on the r side should require 30% more relations, 2100-2200M raw relations. Yield on this test is over 30% higher, and sec/rel is improved by more than 30% also. 35/34 happens to be what CADO would choose for a C210 (they have default settings for every 5 digits, 208 is closest to 210). I'm not convinced this is improved enough to bother trying 35, which is a stretch, but I'd do 34 on both sides for sure!

Wow! Definitely an improvement. I'll point frmky to your posts and he can decide if he wants to up it to 34 or not. Thanks for checking it.

swellman · 2017-12-06, 21:38

QUEUED C231_134_79 is ready for SNFS on 14e.

Code:

n: 676870440131162186790205260284115103229905992405944474763786965158347196908998606158829512914913741518518550185426278771154202748468786511992582732124649322376550585259441177832426707359430632204475614844271700714398827407169704953
# 134^79+79^134, difficulty: 256.41, anorm: 1.83e+039, rnorm: -7.71e+047
# scaled difficulty: 257.85, suggest sieving rational side
# size = 6.549e-013, alpha = 0.000, combined = 8.926e-014, rroots = 0
type: snfs
size: 256
skew: 1.8968
c6: 134
c0: 6241
Y1: -559494740587480879172162808385362976196641
Y0: 4491199828872408503792328704
rlim: 268000000
alim: 450000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 64
rlambda: 2.8
alambda: 2.8

Test sieving on the -r side with Q in blocks of 2K

Code:

20M     4127
80M     2874
150M    2670
250M    2280
350M    1843
400M    1704

Suggesting a sieving range for Q of 20M-410M with target number rels = 480M

RichD · 2017-12-11, 21:41

QUEUED C164 from the OPN t600 file.
(I know I still have one pending but I want to get this listed before it gets lost.)

Code:

n: 15234712737586721103604550217827103691335722219884678818626646892020804317203046936508614696882879218044394957722290829737523608324657357271792501492916421106461453
# 32607907713428723311^13-1, difficulty: 234.16, skewness: 1.00, alpha: 3.10
# cost: 2.41905e+18, est. time: 1151.93 GHz days (not accurate yet!)
skew: 1.000
c6: 1
c5: 1
c4: -5
c3: -4
c2: 6
c1: 3
c0: -1
Y1: -32607907713428723311
Y0: 1063275645447484430688183399163394802722
m: 4116174886578875451779080385257530068405655593255906669725778805326054390667175153667115574583829554496487666898231298223666826336036741546049998074154098460522073
type: snfs
rlim: 67000000
alim: 67000000
lpbr: 30
lpba: 30
mfbr: 60
mfba: 60
rlambda: 2.6
alambda: 2.6

Trial sieving 5K blocks.

Code:

  Q Yield
 20M 9035
 50M 7416
 80M 5811
110M 5749

swellman · 2017-12-12, 12:36

QUEUED C231_135_73 is ready for SNFS on 14e

Code:

n: 405135366248403070014579145301092752429522305796020528619581222136619775733509794280594340666699024601709500245913945891233607180762122241552729694445771745966668715497199198907646172156978215320283919057062908487389773580323648327
# 135^73+73^135, difficulty: 251.55, anorm: 2.01e+032, rnorm: 8.11e+055
# scaled difficulty: 255.48, suggest sieving rational side
# size = 2.418e-017, alpha = 0.000, combined = 8.791e-014, rroots = 1
type: snfs
size: 251
skew: 6.3253
c5: 1
c0: 10125
Y1: -2003521529507672592938232421875
Y0: 204040896602218382792418993938046358519102576817497
rlim: 268000000
alim: 450000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 64
rlambda: 2.8
alambda: 2.8

Test sieving on the -r side with Q in blocks of 2K

Code:

Q=20M    2679
Q=80M    2827
Q=150M   2943
Q=250M   3334
Q=350M   2963

Suggesting a sieving range for Q of 20M-340M with a target # rels = 480M.

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

2017-12-03, 05:30	#1283
VBCurtis "Curtis" Feb 2005 Riverside, CA 2·2,437 Posts	Data on the -a sieving side for the C208: 69/96 -a: 7134 rels, 1.05 sec/rel (35/34) 67/96 -a: 5418 rels, 1.43 sec/rel (34LP) 65-96 -a: 2736 rels, 2.77 sec/rel (33LP) going to 34LP on both sides, even though mfbr is capped at 96, doubles yield and nearly cuts sec/rel in half. Since only 70% more relations are needed, this would result in project length 1.70/1.94 (ratio of extra rels needed to sec/rel ratio), which is around 86%. So, going to 34LP would save 14% of total sieving time. I would aim for 1600-1700M raw relations. If Frmky were interested in running as 34, a couple other Q values should be tested to make sure 34 is always faster. Going to 35LP on the r side should require 30% more relations, 2100-2200M raw relations. Yield on this test is over 30% higher, and sec/rel is improved by more than 30% also. 35/34 happens to be what CADO would choose for a C210 (they have default settings for every 5 digits, 208 is closest to 210). I'm not convinced this is improved enough to bother trying 35, which is a stretch, but I'd do 34 on both sides for sure!