Fast Breeding (guru management) - Page 109

VBCurtis · 2017-10-09, 04:30

Quote:

Originally Posted by swellman

Well, adding 50M additional Q is projected to add ~55M relations for a total of 596M raw relations. And from the recent findings of VBCurtis that won’t be near enough for a 14e/33 job. Suggest Q be bumped up to 550M. From there we can bump it up in smaller increments until the matrix builds successfully at TD=70.

Or just bump Q up to 500M if there’s a risk of overshooting the minimum number of relations. I’ll keep at it until we build a good matrix. But 596M raw relations isn’t going to do it I think.

I think this project is tougher than my 14e/33 effort, so you might need 680-700M raw relations to get a TD 120+ matrix. My job built at TD 140 with 670M raw relations, well below my expectations. I didn't try to build with fewer than 670M raw (540M unique).

swellman · 2017-10-12, 14:03

Quote:

Originally Posted by VBCurtis

I think this project is tougher than my 14e/33 effort, so you might need 680-700M raw relations to get a TD 120+ matrix. My job built at TD 140 with 670M raw relations, well below my expectations. I didn't try to build with fewer than 670M raw (540M unique).

A few days ago I downloaded the relations set, just over 650M relations at that time. Was just looking to get a datapoint on whether msieve could even build a matrix @TD=70 (with the expectation being no). Started msieve on Monday afternoon with this dataset, thinking I’d see the answer when I woke up.

Now three days later and msieve is still chewing. Seems to be stuck on full merge after finding 0-7+ ideals, 2-way merge, then “reduced to 87474860 relation sets and 86454781 unique ideals”, and “26 oversized relation sets”. Presumably it’s low on memory - on a 16 Gb machine. I’ve hit this situation before on a large 15e/32 job. That took over a week to (eventually) build the matrix, along with a ridiculous ETA.

So that’s the status. Even if msieve manages to build a viable TD=70 matrix here, it’s fairly worthless other than data on minimum number of relations required for a 14e/33 job. And I’m wondering if I can complete this job - this is using my second best machine and it’s performance is in the mud. My best (32 Gb) machine will free in mid-January once a 15e/33 job is finished.

FWIW. I’ll just wait until msieve can get through the full merge step and see if a matrix is built.

VBCurtis · 2017-10-12, 16:52

Interesting! My 14/33 job (running on SSD and 16gb) took about 8 hours to complete each instance of -nc1. But I did remdups4 before I tried filtering, which took a couple hours.

As for minimum relations data, I accidentally set a GNFS172 project on 15e/33 rather than the intended 15e/32; it built a TD 104 matrix after just 529M raw relations. This matrix was 9.0M in size, about expected size for GNFS-172. So, if I send a GNFS-173+ to the 14e queue, I'll send it as 33LP to see how the duplicate rate compares to my 15e/33 data point.

Edit: I also ran 13*2^864-1 (SNFS 263-ish) as 15e/33. That built a TD 124 matrix after 618M raw relations, size 15.8M. So, your 650M relation data set *might* build a usable TD 120 matrix, though it's likely to be *big*.

swellman · 2017-10-12, 19:02

Well, NFS@Home is closing on 698M raw relations for C202_137_75. Sounds like it might be enough to build a TD 120-130 matrix, almost certainly at lower TD. Hopefully we’ll know more in a few days.

swellman · 2017-10-12, 21:14

QUEUED C228_146_66 is ready for SNFS as a 14e/32 job.

Code:

n: 403322615866244971015317252029405207385883002450332874986241076661803461639902341122240548701970807294355395451809659292471104327113556821614657457353451021631933889729329464836801059879643933669408128486891525224194947154267597
# 146^66+66^146, difficulty: 245.79, anorm: 1.32e+038, rnorm: -4.58e+046
# scaled difficulty: 247.21, suggest sieving rational side
# size = 3.896e-012, alpha = 0.000, combined = 3.223e-013, rroots = 0
type: snfs
size: 245
skew: 4.0412
c6: 1
c0: 4356
Y1: -22788959716066127411702810573964669100032
Y0: 313726685568359708377
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 in blocks of 2K

Code:

Q=20M    6375
Q=80M    3726
Q=150M   3311
Q=250M   2795
Q=300M   2600

Suggesting a range for Q of 20M-290M with target # rels = 480M.

RichD · 2017-10-14, 21:47

QUEUED C169 from the OPN t600 file.
Sieve on the algebraic side.

Code:

n: 6097289438711591127192360365762092054453165416495578597850906208624110004316023950433701371920699269920504667155602550087593939166837800663343603937040353385051250033689
# 2158316902961^17-1, difficulty: 222.01, skewness: 113.68, alpha: 0.00
# cost: 9.07895e+17, est. time: 432.33 GHz days (not accurate yet!)
lss: 0
skew: 113.680
c6: 1
c0: -2158316902961
Y1: -1
Y0: 10054156379241985814632159892175329681
m: 10054156379241985814632159892175329681
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7
type: snfs

Trial sieving 5K blocks.

Code:

swellman · 2017-10-17, 10:35

QUEUED Yoyo@Home recently got a p51 ECM hit on xyyx composite C229_128_87, leaving a C179 stub. Here is the SNFS poly for 14e.

Code:

n: 14647401704391583862307253743766974488929088078850910189463365031693758011327875999764099304820640355845120750370808754593361829848559861073833136967005304195920625140036050596033
# 128^87+87^128, difficulty: 252.14, anorm: 3.15e+040, rnorm: 4.75e+046
# scaled difficulty: 253.17, suggest sieving rational side
# size = 9.053e-013, alpha = 0.000, combined = 1.153e-013, rroots = 0
type: snfs
size: 252
skew: 3.1332
c6: 8
c0: 7569
Y1: 53691347665120761247143202316447077077687
Y0: -2535301200456458802993406410752
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7

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

Code:

Q=20M    3081
Q=80M    1907
Q=150M   1720
Q=250M   1510
Q=300M   1375

Suggesting a sieving range for Q of 20M-280M for a target # rels = 240M.

RichD · 2017-10-18, 08:19

QUEUED C167 from the OPN t600 file.
Sieve on the algebraic side.

Code:

n: 23327260711114199533184236869267853928958213099573852918831524904354035666636516779708187282668886005285499828361655214369428846201235359235214438370770257933482387771
# 6228362269^23-1, difficulty: 235.06, skewness: 42.89, alpha: 0.00
# cost: 2.59912e+18, est. time: 1237.68 GHz days (not accurate yet!)
lss: 0
skew: 42.894
c6: 1
c0: -6228362269
Y1: -1
Y0: 1504857788884562813973901378315425946321
m: 1504857788884562813973901378315425946321
type: snfs
rlim: 200000000
alim: 200000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7

Trial sieving 5K blocks.

Code:

swellman · 2017-10-18, 08:42

QUEUED (to 15e)

Quote:

Originally Posted by swellman

C226_127_106 is ready for SNFS. Another possible -r and -a simultaneous sieving job? -NO

Code:

n: 2280139691439976870568969462338718473449233379612063824087464058088350308991326968788088698275254458880939735465798971790789201177419031841268155618018789135288844404141201471594727605279414772263530156097952068024055709191881
# 127^106+106^127, difficulty: 257.21, anorm: 2.62e+039, rnorm: -1.12e+049
# scaled difficulty: 258.82, suggest sieving rational side
# size = 1.785e-013, alpha = 1.262, combined = 3.445e-014, rroots = 0
type: snfs
size: 257
skew: 10.9350
c6: 1
c0: 1709674
Y1: -3399563600545615415795972563652626094227456
Y0: 73869809188743794269800200736680064769
rlim: 268000000
alim: 268000000
lpbr: 32
lpba: 32
mfbr: 64
mfba: 64
rlambda: 2.8
alambda: 2.8

If this job is judged to be too much for 14e/32, even if sieved on both sides simultaneously, then 15e/32 appears to work. Test sieving results on the -r side with Q in blocks of 2k.

Code:

Q=20M    rels=4691
Q=80M    rels=3465
Q=150M   rels=3141
Q=250M   rels=2994
Q=350M   rels=2549

Suggesting a sieving range of 20M-350M for Q with a target # of relations=520M.

This job appears to be too cumbersome for another split 14e/32 job sieved on both sides. Suggest it be enqueued as a straightforward 15e/32 job.

swellman · 2017-10-18, 19:51

QUEUED C229_150_58 is ready for SNFS on 14e.

Code:

n: 2278938577267948440576205638759394291619090321721363932624743095904290280575130508313664426833988199618531605192119784698293356275291256402864532690909748774313973256892089290964813190056727674607759428605802289408954085632654453
# 150^58+58^150, difficulty: 248.26, anorm: 3.60e+038, rnorm: 1.00e+047
# scaled difficulty: 249.67, suggest sieving rational side
# size = 1.961e-012, alpha = 0.000, combined = 1.976e-013, rroots = 0
type: snfs
size: 248
skew: 5.6462
c6: 1
c0: 32400
Y1: -375423431396484375
Y0: 237919412134036023751018299488621739966464
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 62
mfba: 62
rlambda: 2.7
alambda: 2.7

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

Code:

Q=20M    3433
Q=80M    2093
Q=150M   1902
Q=250M   1740

Suggesting a sieving range for Q of 20M-250M with a target # of rels = 245M.

RichD · 2017-10-19, 03:33

QUEUED C192 from OPN t550 file.
Sieve on the algebraic side.

Code:

n: 342111864524063293825336262088484226391971928687135843134784616304200123278308375976499881378750510347809914915426740304278934944109297042523531978229423750026979762755763230118223943547133317
# 1577431459^23-1, difficulty: 220.75, skewness: 34.12, alpha: 0.00
# cost: 8.1842e+17, est. time: 389.72 GHz days (not accurate yet!)
lss: 0
skew: 34.119
c6: 1
c0: -1577431459
Y1: -1
Y0: 6191587163130663482001569147410679761
m: 6191587163130663482001569147410679761
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:

2017-10-12, 16:52	#1191
VBCurtis "Curtis" Feb 2005 Riverside, CA 2×2,437 Posts	Interesting! My 14/33 job (running on SSD and 16gb) took about 8 hours to complete each instance of -nc1. But I did remdups4 before I tried filtering, which took a couple hours. As for minimum relations data, I accidentally set a GNFS172 project on 15e/33 rather than the intended 15e/32; it built a TD 104 matrix after just 529M raw relations. This matrix was 9.0M in size, about expected size for GNFS-172. So, if I send a GNFS-173+ to the 14e queue, I'll send it as 33LP to see how the duplicate rate compares to my 15e/33 data point. Edit: I also ran 132^864-1 (SNFS 263-ish) as 15e/33. That built a TD 124 matrix after 618M raw relations, size 15.8M. So, your 650M relation data set might* build a usable TD 120 matrix, though it's likely to be big. Last fiddled with by VBCurtis on 2017-10-12 at 17:04

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

2017-10-12, 19:02	#1192
swellman Jun 2012 11×281 Posts	Well, NFS@Home is closing on 698M raw relations for C202_137_75. Sounds like it might be enough to build a TD 120-130 matrix, almost certainly at lower TD. Hopefully we’ll know more in a few days.