mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   NFS@Home (https://www.mersenneforum.org/forumdisplay.php?f=98)
-   -   2022 Queue management of 15e (https://www.mersenneforum.org/showthread.php?t=27462)

charybdis 2022-01-28 01:51

[QUOTE=RichD;598936]This is a 31/33 hybrid job. I don't know the proper amount of raw relations so I assumed it would be equivalent to a 32/32 job.[/QUOTE]

It will likely be ~20% more than a 32/32 job. I suspect 31/33 is not optimal when you account for this.

RichD 2022-01-28 02:28

[QUOTE=charybdis;598937]It will likely be ~20% more than a 32/32 job. I suspect 31/33 is not optimal when you account for this.[/QUOTE]
I was getting a 20% increase in runtime (sec/rel) at a low Q with only a slight increase in yield thinking this would be better than a 32/33 job which would need more relations. I need to run the full gamut. Please hold off until further testing.

swellman 2022-01-28 02:48

[quote]This is a 31/33 hybrid job. I don't know the proper amount of raw relations so I assumed it would be equivalent to a 32/32 job. This is estimated to be around 435M total relations. Feel free to adjust accordingly.[/quote]

I use the average of target relations for each lpb plus a bit, in this case (260M+930M)/2 = 595M ~ 620M+ (I target 650M but I tend go heavy on the # of relations).

VBCurtis 2022-01-28 03:32

[QUOTE=swellman;598942]I use the average of target relations for each lpb plus a bit, in this case (260M+930M)/2 = 595M ~ 620M+ (I target 650M but I tend go heavy on the # of relations).[/QUOTE]

I think it should be a geometric average, which would land not-much-higher than a 32/32 job. If 440M would do for 32/32, something like 475M should do for this 31/33. That said, I don't think 31/33 is going to be faster for a sextic. You could run mfbr at 62 but lpbr at 32, which should reduce the number of raw relations needed by a dozen million or two- say, to maybe 620M instead of 640M.

charybdis 2022-01-28 13:42

[QUOTE=VBCurtis;598944]I think it should be a geometric average, which would land not-much-higher than a 32/32 job.[/QUOTE]

I've only ever run one sizeable job with lpbs skewed by 2, which was a difficulty-229 quartic with lpb 32/30, but I remember needing substantially more relations than I'd initially expected - closer to the arithmetic mean than the geometric mean. In hindsight I shouldn't have been surprised: the number of possible ideals with 32/30 is roughly the arithmetic mean of 30/30 and 32/32, and while this isn't a perfect predictor of the number of relations, it should at least scale about the same.

swellman 2022-02-03 13:27

[B]QUEUED AS 5m3_425[/B]

5-3,425 is another HCN quartic SNFS of difficulty 237.6. It has survived 32,247 curves of ECM @B1=260M and is ready for SNFS.

[code]
n: 3615585585196150831178452864134253581298092586172168742490880669239890638015724859218917122864482115049097469572394129723238144231684858711215525897319073505719861946180872195895735426929275351
skew: 1.0000
type: snfs
size: 237
c4: 1
c3: 1
c2: 1
c1: 1
c0: 1
Y1: -35917545547686059365808220080151141317043
Y0: 258493941422821148397315216271863391739316284656524658203125
rlim: 134000000
alim: 266000000
lpbr: 33
lpba: 31
mfbr: 96
mfba: 62
rlambda: 3.7
alambda: 2.7
[/code]

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

[code]
MQ Norm_yield Speed (sec/rel)
35 2366 0.400
65 2856 0.381
100 2936 0.391
150 2993 0.397
200 2780 0.451
250 2734 0.432
[/code]

Suggesting a sieving range of 35-260M to generate 635M raw relations.

VBCurtis 2022-02-04 06:06

[QUOTE=VBCurtis;598815]May I have another 15MQ on 13_2_922m1? 600M raw relations looks a little light; I've been targeting 620-630M.

[B]Done.[/B][/QUOTE]

619M relations got me a 31.7M matrix, 3-4M dimensions bigger than similar recent jobs. This job is a few SNFS digits bigger, so I'll call that good 'nuff and get to solving. This one can be moved to postprocessing.

unconnected 2022-02-19 01:27

[B]QUEUED AS C191_933436_12595[/B]

C191 from 933436:i12595 for 15e queue.
Thanks [B]Gimarel[/B] for the ploy!
[CODE]n: 95943792139492908765333071790605979163835027624974805898350717773405577196123295473856044969264458190095842686539462763700839439312300013661751809241655826533008542372113621303929192489690417
# norm 1.457307e-18 alpha -8.343892 e 1.949e-14 rroots 5
skew: 206761148.35
c0: 18645184046033077167247177836639821624045365940
c1: 670519807821465463730496628562844466976
c2: -18498861485056615649666134341771
c3: -170301411285433552736437
c4: 453902219052530
c5: 277200
Y0: -14724839720475995757223413748323367213
Y1: 15912761554705719950713
rlim: 268000000
alim: 134000000
lpbr: 32
lpba: 33
mfbr: 64
mfba: 96
rlambda: 2.8
alambda: 3.7
type: gnfs
lss: 0[/CODE]
Suggesting sieve range is 40M-300M.

RichD 2022-03-03 22:34

[B]QUEUED AS 331_109m1[/B]

C255 from the OPN t550 file.
Second attempt to optimize the parameters.
This should run a bit faster though the special-Q range is about the same.
Estimate of 625M raw relations.
[CODE]n: 456144099782224315696824095020150332081139148797304856352191816675514369613259984102145549799476807322037325051749567250153209607473297001774552409742710930294846495418796912917131669944275190656583437621142715383063574336648632009217696108548318826496061
# 331^109-1, difficulty: 274.66, skewness: 0.38, alpha: 0.00
# cost: 5.17803e+19, est. time: 24657.29 GHz days (not accurate yet!)
skew: 0.380
c6: 331
c0: -1
Y1: -1
Y0: 2274593993441843779293942507658183948042259641
type: snfs
rlim: 134000000
alim: 268000000
lpbr: 32
lpba: 33
mfbr: 64
mfba: 96
rlambda: 2.7
alambda: 3.4[/CODE]Increasing lambda will increase the yield by 1-2% but add 2-4% to timing.

Trial sieving 5K blocks.
[CODE] Q Yield
40M 15514
70M 14241
100M 14472
200M 12038
280M 10788[/CODE]

swellman 2022-03-04 19:16

3+2,625
 
[B]QUEUED AS 3p2_625[/B]

3+2,625 is another HCN which has finished ECM. It has a quartic SNFS polynomial with difficulty 238.

[code]
n: 1350872366946090935056059296879491581002005646059700802359454035513008730498298808685203195085142436073615149011559333058816162978238818959084242569301395890887217531261929477924397448314567011799166352011282568751
type: snfs
size: 238
skew: 1.0000
c4: 1
c3: -1
c2: 1
c1: -1
c0: 1
Y1: -42535295865117307932921825928971026432
Y0: 436673502879206784130402698570834024654748577491697818855443
rlim: 134000000
alim: 266000000
lpbr: 33
lpba: 31
mfbr: 96
mfba: 62
rlambda: 3.7
alambda: 2.7
[/code]

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

[code]
MQ Norm_yield Speed (sec/rel)
40 2350 0.406
70 2855 0.371
100 2749 0.418
150 2866 0.414
200 2630 0.447
250 2445 0.441
300 2611 0.414
[/code]

Suggesting a sieving range for Q of 40-280M to generate 640M raw relations.

swellman 2022-03-07 13:19

[B]QUEUED AS 9m5_625M [/B]

9-5,625M is a HCN which has finished ECM and is now ready for NFS. It is a quartic SNFS of difficulty 238, and should be sieved on the rational side using 15e.

[code]
n: 36360291795869945870161648600451934503491829289459342379822806979996844112392800728837229871877907393616438168811949664371251367603949464962031840484992514841107766422315320491625665424163725530585973435284509712227587890122410608365891251
type: snfs
size: 238
skew: 1.0000
c4: 1
c3: 5
c2: 15
c1: 25
c0: 25
Y1: 21684043449710088680149056017398834228515625
Y0: -436673502879206784130402698570834024654748577491697818855443
rlim: 134000000
alim: 266000000
lpbr: 33
lpba: 31
mfbr: 96
mfba: 62
rlambda: 3.7
alambda: 2.7
[/code]

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

[code]
MQ Norm_yield Speed (sec/rel)
35 1812 0.499
40 2057 0.439
70 2238 0.447
100 2375 0.462
150 2442 0.476
200 2319 0.452
250 2159 0.489
300 2059 0.509
320 2058 0.511
[/code]

Suggesting a sieving range for Q of 36-320M to generate 640M raw relations.


All times are UTC. The time now is 16:29.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, Jelsoft Enterprises Ltd.