2022 Queue management of 15e - Page 13

swellman · 2022-10-18, 12:06

See post 143 below for final polynomial selected for sieving

12+7,256 is a SNFS 276 composite from the HCN project. It is ready for sieving on 15e.

I test sieved this job twice, first as a 33/32 job and then as a 33/33.

33/32 case:

Code:

n: 14465784763601126977599055269783270532544641304323145978534145654240633858626527220916188279798148826301966069344878797378582866606395998179336371169145932268467416189620924711821304788176380624897
skew: 0.2992
type: snfs
size: 276
c6: 12544
c0: 9
Y1: -2183814375991796599109312252753832343
Y0: 6349413173626453466517503997664984086853189632
rlim: 134000000
alim: 266000000
lpbr: 33
lpba: 32
mfbr: 96
mfba: 64
rlambda: 3.6
alambda: 2.4

Results of test sieving on the rational side with Q in blocks of 1000:

Code:

MQ       Norm_yield
35          2457
50          2441
75          2240
100         2030
150         1756
200         1648
250         1458
300         1382
350         1293
400         1268
450         1211

Suggesting a sieving range for Q of 35-475M to generate 700M raw relations.

Test sieved again as a 33/33 job:

Code:

n: 14465784763601126977599055269783270532544641304323145978534145654240633858626527220916188279798148826301966069344878797378582866606395998179336371169145932268467416189620924711821304788176380624897
skew: 0.2992
type: snfs
size: 276
c6: 12544
c0: 9
Y1: -2183814375991796599109312252753832343
Y0: 6349413173626453466517503997664984086853189632
rlim: 134000000
alim: 266000000
lpbr: 33
lpba: 33
mfbr: 96
mfba: 66
rlambda: 3.6
alambda: 2.4

Results of this test sieving, again on the rational side with Q in blocks of 1000:

Code:

MQ       Norm_yield
35          3414
50          3429
75          3137
100         2881
150         2506
200         2317
250         2035
300         1947
350         1817
400         1775
450         1686

Suggesting a sieving range for Q of 35-450M to generate 940M raw relations.

As the 33/33 configuration used 25M less Q and had a healthier yield, I'm going with this second job file.

jyb · 2022-10-20, 16:38

Quote:

Originally Posted by swellman

12+7,256 is a SNFS 276 composite from the HCN project. It is ready for sieving on 15e.

I test sieved this job twice, first as a 33/32 job and then as a 33/33.

33/32 case:

Code:

n: 14465784763601126977599055269783270532544641304323145978534145654240633858626527220916188279798148826301966069344878797378582866606395998179336371169145932268467416189620924711821304788176380624897
skew: 0.2992
type: snfs
size: 276
c6: 12544
c0: 9
Y1: -2183814375991796599109312252753832343
Y0: 6349413173626453466517503997664984086853189632
rlim: 134000000
alim: 266000000
lpbr: 33
lpba: 32
mfbr: 96
mfba: 64
rlambda: 3.6
alambda: 2.4

[snip]

As the 33/33 configuration used 25M less Q and had a healthier yield, I'm going with this second job file.

Setting aside the question of large prime bounds, I'm surprised at the choice of polynomials: that leading coefficient isn't huge, but it's bigger than it needs to be. Did you look at using either of these?

Code:

c6: 49
c0: 144
Y1: -2183814375991796599109312252753832343
Y0: 25397652694505813866070015990659936347412758528

Code:

c6: 196
c0: 9
Y1: -2183814375991796599109312252753832343
Y0: 12698826347252906933035007995329968173706379264

The latter of these has an identical SNFS difficulty to yours, the former's is slightly higher but may nonetheless turn out better with the smaller coefficient.

swellman · 2022-10-20, 16:53

Quote:

Originally Posted by jyb

Setting aside the question of large prime bounds, I'm surprised at the choice of polynomials: that leading coefficient isn't huge, but it's bigger than it needs to be. Did you look at using either of these?

Code:

c6: 49
c0: 144
Y1: -2183814375991796599109312252753832343
Y0: 25397652694505813866070015990659936347412758528

Code:

c6: 196
c0: 9
Y1: -2183814375991796599109312252753832343
Y0: 12698826347252906933035007995329968173706379264

The latter of these has an identical SNFS difficulty to yours, the former's is slightly higher but may nonetheless turn out better with the smaller coefficient.

The SNFS polynomial was provided by Yafu. Except for obligatory quartics, Yafu runs through dozens, or in some cases, over a hundred polynomials before test sieving the “best” three (two sextics and one quintic). Usually gives good results, guess things were a little messier in this case.

ETA: I’ll be happy to test sieve the other two cases you mention. If it saves 10MQ it’s worth checking.

unconnected · 2022-10-21, 20:59

See posts 137-142 below

C190 from aliquot sequence 785232:i11573 is ready for 15e queue:

Code:

n: 4924615299710155724242041611093792534228929027833793207209717741579479862664806712426989267280154775007779415785841244751833760998628602996813364136776619527437902426854181741931202256338971
Y0: -4816798648132866790605815531348495563
Y1: 1711431772495717393
c0: -6560899350388287964076509886023554467783604240
c1: -124813949570426726344271243947893318876
c2: 5758067944148696689000856644812
c3: -32012217666497014189601
c4: 606195740080942
c5: 1899240
skew: 108308506.39
# size 1.231e-18, alpha -7.526, combined = 1.960e-14 rroots = 3
# expecting poly E from 2.55e-14 to > 2.94e-14
rlim: 266000000
alim: 134000000
lpbr: 32
lpba: 33
mfbr: 64
mfba: 96
rlambda: 2.8
alambda: 3.7
type: gnfs
lss: 0

Thanks RichD for the poly!

Suggesting sieving range is 40M-300M.

VBCurtis · 2022-10-21, 21:39

Quote:

Originally Posted by unconnected

C190 from aliquot sequence 785232:i11573 is ready for 15e queue:

Code:

rlambda: 2.8
alambda: 3.7
type: gnfs
lss: 0

Thanks RichD for the poly!
-
Suggesting sieving range is 40M-300M.

Why are lambdas so high? Specifically, rlambda- with lim that large and 32LP, I think you'd sieve faster at 2.5 without losing relations. See https://mersenneforum.org/showpost.p...&postcount=110 and the dozen or so posts afterward.

unconnected · 2022-10-22, 03:42

Sorry i missed this discussion regarding lambdas. My choose of lambdas based mostly on previous experience of similar-sized jobs (f.e. see here or here).
However I've no objections, let it be rlambda=2.5. As for alambda 96*log(2)/log(134000000)=~3.55 and looks OK. Should I redo test-sieve with new parameters?

VBCurtis · 2022-10-22, 03:50

My tests with too-big lambdas vs a smaller non-sieve-side lambda show no difference in relations found (like 0-50 relations in a test-range that finds 5000), but a useful speed improvement like 10+%.

I wouldn't change the sieve-side lambda without test-sieving, because the effective lim is so much smaller for low Q so the needed lambda is higher for low Q. Maybe next job test 3.7 against 3.6?

unconnected · 2022-10-22, 07:41

Ok, got it.

swellman · 2022-10-22, 12:42

QUEUED AS C190_785232_11573

So to confirm everything, we are enqueuing the following:

Code:

n: 4924615299710155724242041611093792534228929027833793207209717741579479862664806712426989267280154775007779415785841244751833760998628602996813364136776619527437902426854181741931202256338971
Y0: -4816798648132866790605815531348495563
Y1: 1711431772495717393
c0: -6560899350388287964076509886023554467783604240
c1: -124813949570426726344271243947893318876
c2: 5758067944148696689000856644812
c3: -32012217666497014189601
c4: 606195740080942
c5: 1899240
skew: 108308506.39
# size 1.231e-18, alpha -7.526, combined = 1.960e-14 rroots = 3
# expecting poly E from 2.55e-14 to > 2.94e-14
rlim: 266000000
alim: 134000000
lpbr: 32
lpba: 33
mfbr: 64
mfba: 96
rlambda: 2.5
alambda: 3.7
type: gnfs
lss: 0

And sieving over the original stated range of q: 40-300M, correct?

unconnected · 2022-10-22, 13:26

Quote:

Originally Posted by swellman

So to confirm everything, we are enqueuing the following:
<snip>

And sieving over the original stated range of q: 40-300M, correct?

Yes. This should produce 650M raw relations.

swellman · 2022-10-23, 00:14

QUEUED AS 12p7_256

Quote:

Originally Posted by jyb

Setting aside the question of large prime bounds, I'm surprised at the choice of polynomials: that leading coefficient isn't huge, but it's bigger than it needs to be. Did you look at using either of these?

Code:

c6: 49
c0: 144
Y1: -2183814375991796599109312252753832343
Y0: 25397652694505813866070015990659936347412758528

Code:

c6: 196
c0: 9
Y1: -2183814375991796599109312252753832343
Y0: 12698826347252906933035007995329968173706379264

The latter of these has an identical SNFS difficulty to yours, the former's is slightly higher but may nonetheless turn out better with the smaller coefficient.

Ran additional test sieving on 12+7,256 based on @jyb's suggestions, one SNFS polynomial is clearly better than the rest:

Code:

n: 14465784763601126977599055269783270532544641304323145978534145654240633858626527220916188279798148826301966069344878797378582866606395998179336371169145932268467416189620924711821304788176380624897
type: snfs
size: 276
skew: 0.5984
c6: 196
c0: 9
Y1: -2183814375991796599109312252753832343
Y0: 12698826347252906933035007995329968173706379264
rlim: 134000000
alim: 266000000
lpbr: 33
lpba: 33
mfbr: 96
mfba: 66
rlambda: 3.65
alambda: 2.45

Results of test sieving on the rational side with Q in blocks of 1000:

Code:

MQ       Norm_yield
35          3692
50          3708
75          3321
100         2995
150         2659
200         2384
250         2184
300         1955
350         1938
400         1802
450         1780

Suggesting a sieving range for Q of 35-430M to generate 950M raw relations, i.e. 20M less Q than previously submitted. We'll go with the above poly on this job.

Thanks to @jyb for righting the ship.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Queue management for 14e queue	VBCurtis	NFS@Home	170	2023-01-02 15:27
2022 - queue management for 15e_small	swellman	NFS@Home	186	2022-12-29 17:58
Queue management for 16e queue	VBCurtis	NFS@Home	154	2022-12-23 21:35
Queue management for e_small and 15e queues	VBCurtis	NFS@Home	254	2022-01-02 01:59
Improving the queue management.	debrouxl	NFS@Home	10	2018-05-06 21:05

2022-10-22, 03:42	#138
unconnected May 2009 Moscow, Russia 2·7·11·19 Posts	Sorry i missed this discussion regarding lambdas. My choose of lambdas based mostly on previous experience of similar-sized jobs (f.e. see here or here). However I've no objections, let it be rlambda=2.5. As for alambda 96*log(2)/log(134000000)=~3.55 and looks OK. Should I redo test-sieve with new parameters?

2022-10-22, 03:50	#139
VBCurtis "Curtis" Feb 2005 Riverside, CA 5²·229 Posts	My tests with too-big lambdas vs a smaller non-sieve-side lambda show no difference in relations found (like 0-50 relations in a test-range that finds 5000), but a useful speed improvement like 10+%. I wouldn't change the sieve-side lambda without test-sieving, because the effective lim is so much smaller for low Q so the needed lambda is higher for low Q. Maybe next job test 3.7 against 3.6?

2022-10-22, 07:41	#140
unconnected May 2009 Moscow, Russia B6E₁₆ Posts	Ok, got it.