Tom, thank you for the advice.

[QUOTE=fivemack;453538]

880 not pushed - I'm really not happy with three-large-primes-both-sides, for that large a number I would definitely recommend 16e.

[/QUOTE]

Not sure I understand your comment - we have factored C191 on 15e via GNFS before have we not? Or is the polynomial or the parameters causing problems? I'm not questioning your decision here, just trying to understand the limits of 15e so I don't exceed them. Post 880 is reposted below for viewing convenience.


[quote]
[code]
n: 29460893303338144751360360097976743017149046981259832053501450854362438285630845458294329588136961317820466603439061784124252469966169391148524869909406500896547611862071404959591325864761463
# norm 1.222655e-18 alpha -8.479287 e 1.735e-14 rroots 3
skew: 17653518.44
c0: 826319531224681659341264214653314828053252960
c1: 26940857953335994224321141436849154800
c2: -22688220016211567990602767782164
c3: 15607687492217763886884
c4: 54690847245705303
c5: 1286485200
Y0: -1870545265872388890161243166679811449
Y1: 554791760382776548817
rlim: 450000000
alim: 450000000
lpbr: 33
lpba: 33
mfbr: 66
mfba: 66
rlambda: 3.0
alambda: 3.0
[/code]

Test sieving on the -a side for 10K range at various Q starting values were as follows:

Q_start yield
150M 2.17
300M 2.15
450M 1.81
600M 1.80

alim of 450M seems to work nicely, but perhaps it should be 500M. This factorization is outside my comfort zone.
[/quote]

I have pushed C171_829332_3638 to 14e rather than 15e, despite

[code]
> ../gnfs-lasieve4I15e -a C171_829332_3638.gnfs -f 268000000 -c 1000
total yield: 7103, q=268001011 (0.12773 sec/rel)
> ../gnfs-lasieve4I14e -a C171_829332_3638.gnfs -f 268000001 -c 999
total yield: 3104, q=268001011 (0.19458 sec/rel)
[/code]

because the 14e queue was getting quite empty.

[QUOTE=swellman;453593]Not sure I understand your comment - we have factored C191 on 15e via GNFS before have we not? Or is the polynomial or the parameters causing problems? I'm not questioning your decision here, just trying to understand the limits of 15e so I don't exceed them. Post 880 is reposted below for viewing convenience.[/QUOTE]

I made a mistake, sorry; I saw that you had rlambda=alambda=3.0 and thought that this meant you were trying for three large primes on both sides, which in my experience causes moderately higher yields at the price of vastly longer sieving times. But since you didn't have alim or rlim equal to 96, this isn't what you were doing.

Had you in fact considered and rejected using 32-bit large primes with three on one side?

(Maybe I'm being a bit critical recently; I trust the participants on this bit of the forum to tell me if I am in danger of turning into RDS)

[QUOTE=fivemack;453632]I made a mistake, sorry; I saw that you had rlambda=alambda=3.0 and thought that this meant you were trying for three large primes on both sides, which in my experience causes moderately higher yields at the price of vastly longer sieving times. But since you didn't have alim or rlim equal to 96, this isn't what you were doing.

Had you in fact considered and rejected using 32-bit large primes with three on one side?

(Maybe I'm being a bit critical recently; I trust the participants on this bit of the forum to tell me if I am in danger of turning into RDS)[/QUOTE]

No worries, I was just confused on the 3 LP comment. No, I have not played with 3 on one side - not sure how best to do that. Assuming we want to sieve this polynomial on the -a side, would we make lpba=32 and mfba=96? Or do we make the rational side 32/96?

Or I could test sieve it and learn something in the process...nah that sounds hard.:whistle:

[QUOTE=swellman;453635]No worries, I was just confused on the 3 LP comment. No, I have not played with 3 on one side - not sure how best to do that. Assuming we want to sieve this polynomial on the -a side, would we make lpba=32 and mfba=96? Or do we make the rational side 32/96?

Or I could test sieve it and learn something in the process...nah that sounds hard.:whistle:[/QUOTE]

I think you probably do have to test-sieve: I got, where '32A' means 3 primes on algebraic side, 2 on rational side, sieve on algebraic side, so
[code]
lpbr: 32
lpba: 32
mfbr: 64
mfba: 96
rlambda: 2.6
alambda: 3.6

../gnfs-lasieve4I15e -a C191_3408_1668_a3r2 -f 450000000 -c 1000 2> 32A.t
[/code]

[code]
22A.t total yield: 830, q=450001043 (1.15306 sec/rel)
22R.t total yield: 510, q=450001043 (2.17775 sec/rel)
23A.t total yield: 869, q=450001043 (1.22087 sec/rel)
23R.t total yield: 513, q=450001043 (2.08862 sec/rel)
32A.t total yield: 991, q=450001043 (1.16914 sec/rel)
32R.t total yield: 659, q=450001043 (1.90247 sec/rel)
[/code]

So you really want to sieve on the algebraic side, three algebraic-side primes gets somewhat better yield than three rational-side primes, and the time per relation is basically a wash between 3A2R and 2A2R. I've queued up with three algebraic-side primes and run 50M to 500M. The linear algebra will be a bit of a bear, but I have reasonable bear-hunting equipment.

Thank you for the analysis and explanation. I think I now understand how to approach 3LP sieving. (When to consider it is another question for another day.)

And thanks for adding it to the 15e queue.

Which poly did you end up using for C171_829332_3638?

[QUOTE=richs;453662]Which poly did you end up using for C171_829332_3638?[/QUOTE]

This one:
[CODE]
N 143078471890001396404975722954703592875148935188191068938072316014184174902812459038968421963880118593068060667758149846867638636307227896395788079777289956829055499772249
SKEW 12640532.26887
A0 577300829383670786734653724089000359155200
A1 -45090526843823845959056529260126600
A2 -6637850955461852556600436653
A3 111155248878476001023
A4 42917566535268
A5 5418720
R0 -691735415860028161291242601067023
R1 25898420289952162105307
SRLPMAX 4294967296
SALPMAX 4294967296
FAMAX 268000000
FRMAX 268000000
[/CODE]

Two more for 14e
 

Both are finishing t55 courtesy of yoyo@home.

C201_142_80
[code]
n: 120995239996089669083457692866260337688533647337487351938695070105850965556070387470375584471251348523080226151866932692127422627517183789564535374387267346411032355957325431945088067023994808350857161
# 142^80+80^142, difficulty: 248.76, anorm: 1.42e+039, rnorm: 6.08e+046
# scaled difficulty: 250.03, suggest sieving rational side
# size = 8.704e-013, alpha = 0.000, combined = 1.119e-013, rroots = 0
type: snfs
size: 248
skew: 2.8854
c5: 1
c0: 200
Y1: 5902958103587056517120000000000000000000000000000
Y0: -416997623116370028124580469121
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7
[/code]



C202_131_69
[code]
n: 9524599816149157292230178818340396418821256941113740648632660387897518147510730794519251713695587871282984830937364114961224882066422395348373955394386577670494244130032985516715160838388710839910210113
# 131^69+69^131, difficulty: 244.64, anorm: 2.24e+041, rnorm: 3.42e+045
# scaled difficulty: 244.64, suggest sieving algebraic side
# size = 1.134e-012, alpha = 0.000, combined = 1.377e-013, rroots = 0
type: snfs
size: 244
skew: 6.183
c5: 1
c0: 9039
Y1: 645767760190718297889184705014020888804383290681
Y0: -438326915318176225182722457721
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7
[/code]

Out of town for 1.5 weeks and everything is caught up!

Using the previous p^19-1 as a template we now have from the [URL=http://www.lirmm.fr/~ochem/opn/t600.txt]t600[/URL] file.
[CODE]n: 5388792835728955147464475071441990735548637250874361350066071879413146964544778832432006808552504514052955962054124335878565007804542802788136376070261154961282201213602489
# 327668647153^19-1 (C172) sieve on algebraic side
lss: 0
skew: 0.012
c6: 327668647153
c0: -1
Y1: -1
Y0: 35180715207538134081393278607450577
rlim: 67000000
alim: 67000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.6
alambda: 2.6[/CODE]