2022 Queue management of 15e - Page 9

swellman · 2022-07-25, 11:43

QUEUED AS 314__757_19m1

So, for the record, we are enqueuing this with 15e:

Code:

n: 15408601493056043415906090162578650013956154779559962936558235847864676250787149677973037155754572123552662048761330289226141167641708444550944783565733312174411831602455187585645867987694560088894019633038130261626389
# 31479823396757^19-1, difficulty: 256.46, skewness: 0.01, alpha: 0.00
# cost: 1.35513e+19, est. time: 6452.98 GHz days (not accurate yet!)
skew: 139.27
c6: 1
c0: -31479823396757
Y1: 31195852758590190231949248764353333010093
Y0: -1
lss: 0
type: snfs
rlim: 266000000
alim: 134000000
lpbr: 33
lpba: 33
mfbr: 66
mfba: 96
rlambda: 3.0
alambda: 3.7

For Q of 40-450M to start, correct?

ETA: And still sieve on the algebraic side, correct?

charybdis · 2022-07-25, 11:53

Quote:

Originally Posted by RichD

The 300M yield seems to be an outlier, with the expected yield to be part way between 200M and 400M.

Are you normalizing your yields?

Quote:

Would it be a good rule of thumb if the skew is less than, say 0.05, to perform the inversion? Another difficult poly for OPN would be for p^31-1 but not as drastic.

Theoretically it should be better to invert whenever skew is less than 1, but whether this actually matters is a function of how extreme the skew is and how hard the job is: both will tend to make the b values of the relations larger.

This doesn't actually have anything to do with the NFS algorithm itself, only a single limitation of the msieve code. A brief explanation: relations are given by (a,b) pairs, and a typical relation will have a/b of the same order of magnitude as the skew. Msieve is currently limited to b values at most 2^32, but there is no such limit on a. Inverting swaps a with b, so if the skew is less than 1 then inverting should reduce the number of relations thrown out by msieve. Though for skews not too far from 1, this is unlikely to matter much: e.g. this OPN job run on 15e with skew 0.24, where less than 0.01% of the relations were lost. On the other hand, here is one with skew 0.01 where almost 2% of the relations were thrown out. 31479823396757^19-1 is bigger and has an even more extreme skew so the losses would be greater than that.

Quote:

Originally Posted by swellman

So, for the record, we are enqueuing this with 15e:
...

Looks correct to me. The parameters don't need to change when you invert the polynomials, as you're mainly producing the exact same relations with the same norms, the only difference being that a and b have swapped.
You decide whether this counts as a "smaller number" that can go before 2,2862M

Quote:

ETA: And still sieve on the algebraic side, correct?

Yes.

swellman · 2022-07-29, 17:46

QUEUED AS 11p8_550M

11+8,550M is a GNFS 195 composite from the HCN project ready for sieving on 15e.

Code:

n: 136288161210072395823013117019980969502173240018666412020739446651656793905151246496169927562302139212209956899378445233588256353492133274266113370748323615532523317306454641969468409364934537501
skew: 200515477.36
c0: -32242163876522035851665931660453021024510617584
c1: 9073984765321950964906593967605583729980
c2: 6549708452747032989837187045500
c3: -564303215770502934756051
c4: 332286224984506
c5: 4826640
Y0: -30914435748287354232384753627221472783
Y1: 6830632680016625778885277
# MurphyE (Bf=8.590e+09,Bg=4.295e+09,area=2.684e+16) = 9.779e-09
skew 200515477.36, size 4.142e-019, alpha -7.334, combined = 9.980e-015 rroots = 5
rlim: 266000000
alim: 134000000
lpbr: 33
lpba: 33
mfbr: 66
mfba: 96
rlambda: 3.0
alambda: 3.7

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

Code:

MQ      Norm_yield     Speed (sec/rel)
20         1812             0.496
35         2194             0.427
50         2447             0.419
75         2376             0.472
100        2546             0.493
140        2461             0.472
180        2252             0.534
240        2053             0.520
300        1866             0.580
360        1767             0.600
420        1725             0.611
480        1637             0.700

Suggesting a sieving range for Q of 20-490M to generate 950M raw relations.

VBCurtis · 2022-07-30, 01:37

For gnfs jobs, qmax should be no more than 8 times qmin, otherwise the duplicate ratio gets so high as to make the lower-Q relations not worth gathering.
Maybe start this job at Q= 60M? Or is my understanding (gleaned from CADO experiments) incomplete?

swellman · 2022-07-30, 11:39

Quote:

Originally Posted by VBCurtis

For gnfs jobs, qmax should be no more than 8 times qmin, otherwise the duplicate ratio gets so high as to make the lower-Q relations not worth gathering.
Maybe start this job at Q= 60M? Or is my understanding (gleaned from CADO experiments) incomplete?

I can’t speak to best practices in sieving with CADO, but experience has shown ggnfs results do not seem reliant on the qmax/qmin ratio.

RichD · 2022-08-02, 10:59

QUEUED AS 1888171_41m1

C239 from the OPN t600 file.

Code:

n: 96213951613469305231302231894763722298853923706881919967966443770584688870752537981009385374951754089030434734044104610926201077424443533921747920930246549621622346963452693177188213019619789319025026149312773142478891851859089959013724557
# 1888171^41-1, difficulty: 263.59, skewness: 11.12, alpha: 0.00
# cost: 2.3066e+19, est. time: 10983.79 GHz days (not accurate yet!)
skew: 11.117
c6: 1
c0: -1888171
Y1: -1
Y0: 85563641971119684450629651781443451031149091
type: snfs
rlim: 134000000
alim: 134000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 93
rlambda: 2.8
alambda: 3.5

I found out one needs to add the -v switch to the GGNFS sievers to get the additional information for normalization. With a more exact science I can use a smaller sieve interval which will save me tons of time on my C2D laptop.

Trial sieving 2K blocks.

Code:

  Q  Yield N-Yld
 40M  5990  5800
 70M  5282  5467
100M  5013  4948
200M  4250  4118
232M  4388  3829

charybdis · 2022-08-05, 14:05

QUEUED AS 2_1443m1

2,1443- is a Cunningham composite of SNFS difficulty 290 from the recent base 2 extension. Hopefully this will be the last one that I have to send to 15e.

Code:

n: 14916922976842176669639475705469515012190536891037011036656143393536518151666274307635319497442835675117923217375851417102963789797694391510612078590311433370085309252348348391596546545270297055271396478769686564766226428432338972804920032231499823
skew: 1.08
type: snfs
size: 290
c6: 4
c3: 2
c0: 1
Y1: 1
Y0: -1461501637330902918203684832716283019655932542976
rlim: 134000000
alim: 266000000
lpbr: 33
lpba: 33
mfbr: 96
mfba: 66
rlambda: 3.8
alambda: 2.6

Test-sieving 1k ranges on the rational side:

Code:

MQ       Norm_yield      Speed (sec/rel)
25          3093              0.134
40          3008              0.145
70          2643              0.168
100         2512              0.181
150         2013              0.228
200         1828              0.246
300         1567              0.278
400         1508              0.283
500         1329              0.314
600         1255              0.327

Suggest sieving 25-560M for 950M raw relations.

kruoli · 2022-08-08, 17:40

QUEUED AS 3_565p

GCW 3,565+ is ready for sieving.

Code:

n: 3443671211875592673692410405878773741590128830637420629071903338802101552170315523638510310711142179327881242898400825308153469780588367727373519589991041267518970815116866106440572589
skew: 45005263.17
# skew by cownoise
c0: -166073624299973756029181204680717636447622880
c1: 12672586061064920164358430844686067472
c2: 349011108967340959348153651598
c3: -14788811629198192142371
c4: 439757720590086
c5: 3492720
Y0: -250478543763798807115283741750404037
Y1: 24096991062348010806961
# MurphyE (Bf=8.590e+09,Bg=4.295e+09,area=5.469e+16) = 1.867e-08
# f(x) = 3492720*x^5+439757720590086*x^4-14788811629198192142371*x^3+349011108967340959348153651598*x^2+12672586061064920164358430844686067472*x-166073624299973756029181204680717636447622880
# g(x) = 24096991062348010806961*x-250478543763798807115283741750404037
rlim: 266000000
alim: 134000000
lpbr: 31
lpba: 32
mfbr: 62
mfba: 94
rlambda: 2.7
alambda: 3.5

Test sieving 1k ranges and normalizing gave:

Code:

Q	norm. yield
30M	2180
40M	2235
70M	2321
125M	2229
175M	2025
225M	1745
300M	1639

Sieving from 30M-210M on the algebraic side should give around 370M raw relations.

I will do the LA.

kruoli · 2022-08-10, 10:05

QUEUED AS 10_277m

GCW 10,277- (C188 GNFS) is ready for sieving:

Code:

n: 10346906288308418154637357815932916737602636349051602874820552349773463555930995873832540659095439340742928616672122132398600658384185025923279034022311767253405048658347876100244912612609
skew: 6411780.37383
# skew by cownoise
type: gnfs
c0: -1545280909574661931755671438530522953271140
c1: -284528837209326587649391440411066113
c2: 15342760644173924075294924469
c3: 53422542002613561282002
c4: 7444023615806274
c5: -671766480
Y0: -916576447844309587301085488226030971
Y1: 379186263590257755049441
# MurphyE (Bf=8.590e+09,Bg=4.295e+09,area=5.469e+16) = 1.395e-08
# scored 2.82894726e-14 by cownoise
# f(x) = -671766480*x^5+7444023615806274*x^4+53422542002613561282002*x^3+15342760644173924075294924469*x^2-284528837209326587649391440411066113*x-1545280909574661931755671438530522953271140
# g(x) = 379186263590257755049441*x-916576447844309587301085488226030971
rlim: 266000000
alim: 134000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 94
lss: 0
rlambda: 2.8
alambda: 3.5

Test sieving 1k ranges and normalizing gave:

Code:

Q	norm. yield
35M	2128
40M	2121
70M	2165
125M	2119
200M	1864
300M	1546

Sieving from 35M-275M on the algebraic side should give around 470M raw relations.

I will do the LA.

unconnected · 2022-08-14, 09:43

QUEUED AS C186_933436_12612

C186 from 933436:i12612 for 15e queue:

Code:

n: 268122889478028298755689271666314793363252818657940823233461616255623056680956479871977028159422090514406704183590678406934517962980756137458860542018709710214742647672123344227928654081
# norm 3.742754e-18 alpha -7.935221 e 3.438e-14 rroots 5
skew: 207490587.75
c0: 38170573534334446585237756207909423136814307135
c1: 581841713472748961318528209461747286867
c2: -11832725730178702659350046327181
c3: -33384625446095806142251
c4: 314947443047070
c5: 298200
Y0: -978960681695788806147626971164295158
Y1: 9918096082374422917
rlim: 266000000
alim: 134000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 94
rlambda: 2.8
alambda: 3.5
type: gnfs
lss: 0

Suggesting sieve range is 30M-250M, this should produce ~470M raw relations.

pinhodecarlos · 2022-08-21, 06:05

Hey guys, thank you so much for keeping this queue loaded. I need a favour. There are thousands of cores only allocated to this queue who are idle. Can we move one of the 16e smaller numbers (C214_139_87) here taking into consideration that each Wu's should fit under 1GB/thread. Thank you.

2022-07-30, 01:37	#92
VBCurtis "Curtis" Feb 2005 Riverside, CA 3·1,877 Posts	For gnfs jobs, qmax should be no more than 8 times qmin, otherwise the duplicate ratio gets so high as to make the lower-Q relations not worth gathering. Maybe start this job at Q= 60M? Or is my understanding (gleaned from CADO experiments) incomplete? Last fiddled with by VBCurtis on 2022-07-30 at 01:38

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

2022-08-21, 06:05	#99
pinhodecarlos "Carlos Pinho" Oct 2011 Milton Keynes, UK 2³×641 Posts	Hey guys, thank you so much for keeping this queue loaded. I need a favour. There are thousands of cores only allocated to this queue who are idle. Can we move one of the 16e smaller numbers (C214_139_87) here taking into consideration that each Wu's should fit under 1GB/thread. Thank you.