Queue management for 14e queue - Page 3

chris2be8 · 2021-04-30, 16:03

QUEUED AS f29_173m1r and f29_173m1a

(29^173-1)/564781364 from the Brent tables:

Code:

# Built Thu Apr 29 19:01:17 2021
# Estimated SNFS difficulty 255, GNFS equivalent 183, GNFS difficulty 245, degree 6
# 173 is 1 mod 4, 3 mod 5 and 5 mod 6
# Base1 29: 29
# Base2 1:
n: 17497390267872014162132719492965729852377204869710780460625531119888806195294072779174290738197546522248810149597988727035232260424513075064743584364619129975777226861225183652916820767229528880135752399446124400846343472628423215897778346927967
type: snfs
c6: 1
c0: -29
# Y0 = 29^29
Y0: 2567686153161211134561828214731016126483469
# Y1 = 1^0
Y1: 1
# msieve rating: skew 1.75, size 1.762e-12, alpha 0.852, combined = 1.751e-13 rroots = 2
skew: 1.75
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 91
mfba: 62
rlambda: 3.5
alambda: 2.6

I'd normally send it to 15e_small but I'll put it here to give the BOINC pentathlon some work to do.

Sieve on *both* sides from 20M to 134M. That should get a total of about 300M relations which should allow for the higher duplicate rate sieving on both sides.

Test sieving 10k ranges:

Code:

  Q  yield (algebraic)
 20M 17289
 80M 12059
134M 10309
160M 12187
220M 9695

  Q  yield (rational)
 20M 18995
 80M 12100
134M 10883
160M 10266
220M 9140

I'll do the LA. (And test my script for merging relations from two jobs.)

Chris

swellman · 2021-04-30, 18:56

Quote:

Originally Posted by chris2be8

(29^173-1)/564781364 from the Brent tables:

Code:

# Built Thu Apr 29 19:01:17 2021
# Estimated SNFS difficulty 255, GNFS equivalent 183, GNFS difficulty 245, degree 6
# 173 is 1 mod 4, 3 mod 5 and 5 mod 6
# Base1 29: 29
# Base2 1:
n: 17497390267872014162132719492965729852377204869710780460625531119888806195294072779174290738197546522248810149597988727035232260424513075064743584364619129975777226861225183652916820767229528880135752399446124400846343472628423215897778346927967
type: snfs
c6: 1
c0: -29
# Y0 = 29^29
Y0: 2567686153161211134561828214731016126483469
# Y1 = 1^0
Y1: 1
# msieve rating: skew 1.75, size 1.762e-12, alpha 0.852, combined = 1.751e-13 rroots = 2
skew: 1.75
rlim: 134000000
alim: 134000000
lpbr: 31
lpba: 31
mfbr: 91
mfba: 62
rlambda: 3.5
alambda: 2.6

I'd normally send it to 15e_small but I'll put it here to give the BOINC pentathlon some work to do.

Sieve on *both* sides from 20M to 134M. That should get a total of about 300M relations which should allow for the higher duplicate rate sieving on both sides.

Test sieving 10k ranges:

Code:

  Q  yield (algebraic)
 20M 17289
 80M 12059
134M 10309
160M 12187
220M 9695

  Q  yield (rational)
 20M 18995
 80M 12100
134M 10883
160M 10266
220M 9140

I'll do the LA. (And test my script for merging relations from two jobs.)

Chris

Bolding mine.

Obviously this job is larger than the recently adopted paradigm allows, with 31-bit lpb and lims > 100M. But 14d IS running dry and we do need to feed the pentathlon beast. (Plus I like the idea of your script. Could be easier than appending files and then running remdups.)

I’ll enqueue both job files later tonight unless someone objects.

VBCurtis · 2021-04-30, 19:12

I don't understand why we need to feed 14 at all. I thought the idea was to intentionally let it run dry, only feeding it jobs that are clearly inefficient on 15e; when it's dry, more workunits get done on the other sievers that run jobs we care about / are too hard for single-machine factorers.

This isn't one of those jobs, to my eye (I didn't test sieve).

We have quite a backlog on 15e-small, but a horde of BOINCers might fix that up right quick. There's no reason to try to fill 14e jobs to feed this army- the army should simply look to larger sievers.

I don't object to running this job as submitted, but I wish to discourage future submitters from using 14 when 15e-small is as efficient for their job.

RichD · 2021-05-01, 00:28

Quote:

Originally Posted by VBCurtis

There's no reason to try to fill 14e jobs to feed this army- the army should simply look to larger sievers.

I don't object to running this job as submitted, but I wish to discourage future submitters from using 14 when 15e-small is as efficient for their job.

So is there a lower limit for 15e jobs? I have plenty that would work with 15e/28-bit. Assuming they would run faster on 15e than 14e my C2D laptop could complete one in under two weeks. I have over 50 candidates in that range (not counting GNFS). And I still need to finish the ECM work.

I know this does nothing to advance the knowledge to move the wavefront forward and upward. But offering a few small numbers here or there to welcome newcomers to be introduced into integer factoring techniques would be beneficial for the generations to come. I'm not sure every BOINCer has 8GB to devote to their 8/10/12 thread box. Just my two cents.

VBCurtis · 2021-05-01, 07:35

I think the lower bound for e-small is any job that is no slower on 15e than on 14e. Jobs that are clearly faster on 14e should still go there (until the 14e queue disappears entirely, but I think that's a while away).

In practice, there's a small slice of jobs that are pretty long for solo factorers but not faster on 15e- something like GNFS 156-166 in size (SNFS resists clean size examples). If there aren't any of those to do for some time, we shouldn't try to scare up work- the clients should just bounce to the other queues, and e-small client uses very little memory now that we restrict lim's to 134M.

My objection to this specific job was poster's admission that he was sending it to 14e solely because of the possible BOINC army, and that it would be faster on 15e-small. He does get to test new scripts, thus my attempt to say "let's not make a habit of this" rather than "let's not do this."

chris2be8 · 2021-05-01, 16:02

I won't try to send any more jobs this big to 14e. It may well take twice as long to solve the matrix as it would on 15e_small (using 3LP makes the matrix bigger).

As I said it's only to give the BOINC army something to do.

Chris

chris2be8 · 2021-05-05, 08:04

I havn't enough relations to build a decent matrix, I underestimated how many duplicates it would have. From remdups:

Code:

Found 178553883 unique, 108375842 duplicate (37.8% of total), and 422 bad relations.

Please add another 10M special-Q on each side. That should be enough to build a decent matrix. At least I now know how many duplicates sieving on both sides generates.

And I should have timed it to start sieving when/if the BOINC crowd were about to start. Sorry.

Chris

PS. I'll do f13_227p1 first. So no rush to add relations.

RichD · 2021-05-15, 23:31

QUEUED AS 87777_231

C157 from the NearRepdigit (87777_231) project.

Code:

n: 2554492351743203553466465047163590561279014056908519070000000685201516774421702793327336915507089727958089217742945779099282127536202744456448278526479422651
lss: 0
Y0: -2530893774868872527835187432854
Y1: 11653837865472407
c0: -39861548823757403003701316883289116185
c1: -57577200590567430222475767684207
c2: 49851357607445266131094441
c3: -11633836605299727549
c4: -1087030755260
c5: 24600
skew: 7524172.89
# size 2.933e-15, alpha -6.955, combined = 2.123e-12 rroots = 3
# expecting poly E from 2.07e-12 to > 2.38e-12
type: gnfs
rlim: 33500000
alim: 33500000
lpbr: 29
lpba: 29
mfbr: 58
mfba: 84
rlambda: 2.5
alambda: 3.4

Trial sieving 5K blocks.

Code:

unconnected · 2021-05-26, 14:56

QUEUED AS C162_933436_12583

C162 from 933436:i12583 is ready for 14e queue:

Code:

n: 492814440383158169300421235157104557020141961710096944800293608770273262772737660322494789605417517244000144496936368384289906386354399863132915413818334927538079
#best poly = # norm 9.896824e-16 alpha -7.861673 e 9.968e-13 rroots 5
skew: 23970823.73
c0: 2860257410783843537547046544643972737875
c1: 25887191519919336129637313136675905
c2: -1954268231128157916500821231
c3: -150432834073295874569
c4: 2932594691736
c5: 62220
Y0: -23974559707028278928079194033688
Y1: 255021438453302381
rlim: 67000000
alim: 67000000
lpbr: 30
lpba: 30
mfbr: 60
mfba: 60
rlambda: 2.6
alambda: 2.6
type: gnfs
lss: 0

Suggesting sieve range is 10M-80M, I'll take the LA.

unconnected · 2021-06-05, 19:17

QUEUED AS C156_933436_12587

C156 from 933436:i12587 is ready for 14e queue:

Code:

n: 393069405426520546041165552487213463994423501092845882147090913579845626474402997372468369111016683966614281891830133145404734278919600915014678101311164197
# norm 4.653728e-15 alpha -7.484903 e 2.469e-12 rroots 5
skew: 5818513.15
c0: -238178735264108614403807688709958006680
c1: -18469353974337866776878533768214
c2: 73296910773490961770168490
c3: 9411861146553870327
c4: -2697165600778
c5: 17940
Y0: -1854085572812581669459449455649
Y1: 58769492572130923
rlim: 33500000
alim: 33500000
lpbr: 29
lpba: 29
mfbr: 58
mfba: 58
rlambda: 2.6
alambda: 2.6
type: gnfs
lss: 0

Suggesting sieve range is 10M-50M, I'll take the LA.

unconnected · 2021-06-16, 13:21

QUEUED AS C164_11040_10195

C164 from 11040:i10195 for 14e queue:

Code:

n: 42541643393856956284903801374153022485745789688090735805006839970929485543176874031400528561921598368494282366267619297281575951112040249113647989117082870117895407
# norm 7.976786e-16 alpha -7.285364 e 8.869e-13 rroots 3
skew: 11168228.27
c0: 9763186207486878360027080955455829974100
c1: 4608639415241852800259226388856940
c2: 365209133197998245515112405
c3: -45561072169562837834
c4: -4070906128960
c5: 225456
Y0: -45201255544227030420330166057961
Y1: 150568046173107869
rlim: 67000000
alim: 67000000
lpbr: 30
lpba: 30
mfbr: 60
mfba: 60
rlambda: 2.6
alambda: 2.6
type: gnfs
lss: 0

Suggesting sieve range is 10M-90M, I'll take the LA.

2021-04-30, 19:12	#25
VBCurtis "Curtis" Feb 2005 Riverside, CA 13·433 Posts	I don't understand why we need to feed 14 at all. I thought the idea was to intentionally let it run dry, only feeding it jobs that are clearly inefficient on 15e; when it's dry, more workunits get done on the other sievers that run jobs we care about / are too hard for single-machine factorers. This isn't one of those jobs, to my eye (I didn't test sieve). We have quite a backlog on 15e-small, but a horde of BOINCers might fix that up right quick. There's no reason to try to fill 14e jobs to feed this army- the army should simply look to larger sievers. I don't object to running this job as submitted, but I wish to discourage future submitters from using 14 when 15e-small is as efficient for their job. Last fiddled with by VBCurtis on 2021-04-30 at 19:12

2021-05-01, 07:35	#27
VBCurtis "Curtis" Feb 2005 Riverside, CA 13·433 Posts	I think the lower bound for e-small is any job that is no slower on 15e than on 14e. Jobs that are clearly faster on 14e should still go there (until the 14e queue disappears entirely, but I think that's a while away). In practice, there's a small slice of jobs that are pretty long for solo factorers but not faster on 15e- something like GNFS 156-166 in size (SNFS resists clean size examples). If there aren't any of those to do for some time, we shouldn't try to scare up work- the clients should just bounce to the other queues, and e-small client uses very little memory now that we restrict lim's to 134M. My objection to this specific job was poster's admission that he was sending it to 14e solely because of the possible BOINC army, and that it would be faster on 15e-small. He does get to test new scripts, thus my attempt to say "let's not make a habit of this" rather than "let's not do this." Last fiddled with by VBCurtis on 2021-05-01 at 07:36

2021-05-05, 08:04	#29
chris2be8 Sep 2009 2425₁₀ Posts	I havn't enough relations to build a decent matrix, I underestimated how many duplicates it would have. From remdups: Code: Found 178553883 unique, 108375842 duplicate (37.8% of total), and 422 bad relations. Please add another 10M special-Q on each side. That should be enough to build a decent matrix. At least I now know how many duplicates sieving on both sides generates. And I should have timed it to start sieving when/if the BOINC crowd were about to start. Sorry. Chris PS. I'll do f13_227p1 first. So no rush to add relations. Last fiddled with by chris2be8 on 2021-05-05 at 08:08 Reason: Added PS.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Queue management for e_small and 15e queues	VBCurtis	NFS@Home	254	2022-01-02 01:59
System management notes	kriesel	kriesel	10	2021-09-01 15:56
Run down the queue on MPRIME without quitting GIMPS	Rodrigo	Software	7	2018-05-25 13:26
Improving the queue management.	debrouxl	NFS@Home	10	2018-05-06 21:05
split a prime95 queue & client installation	joblack	Information & Answers	1	2009-01-06 08:45

2021-05-01, 16:02	#28
chris2be8 Sep 2009 2425₁₀ Posts	I won't try to send any more jobs this big to 14e. It may well take twice as long to solve the matrix as it would on 15e_small (using 3LP makes the matrix bigger). As I said it's only to give the BOINC army something to do. Chris