Improved params files, 2022 edition

[swellman]

Originally Posted by VBCurtis

I won't have machines powered on & time to sit down until the end of August.

No worries.

How about a baseline run with the C165 params from the 14e-queue thread, but run on e-small? (not CADO)
That gives us a starting point from which I can try other settings on future jobs.

Specifically:

rlim: 50000000
alim: 40000000
lpbr: 31
lpba: 32
mfbr: 60
mfba: 90
rlambda: 2.4
alambda: 3.55

Sure, we can try it on 15e_small. That queue always needs feeding! It will take a few days for the poly search but it won’t be a problem.

If 15e yields so high that 275M relations arrives before 45M, drop alim to 33.5M.

Noted. (I presume you mean 15e_small here.) I will likely start the Q range as 14-45M to see how yield does, but isn’t this awfully optimistic based on the sieving range used in post 80?

You might also try 14e, with rlim increased to 67M and target_rels 15M higher.

We can do that.

[VBCurtis]

Originally Posted by swellman

Noted. (I presume you mean 15e_small here.) I will likely start the Q range as 14-45M to see how yield does, but isn’t this awfully optimistic based on the sieving range used in post 80?

I meant 15e as the siever we use to test-sieve; that's not different from e-small.

I expect Q=10-43M to be about enough, but 14-48M might prove faster. Going up a siever generally means a sieve range of less than half, so 68MQ on 14e might become 30-32MQ on 15e.

[swellman]

CADO produced a decent poly.

n: 167557249511653871458266357551546863458885530690570804232412193290363836755090460398519979587848735731529759811792409221300973956103572825999817801286921797976146343967
skew: 10405411.22
type: gnfs
lss: 0
c0: 3661678253336036003862651470385669302928
c1: 835118785426414901889382598987474
c2: -4036779104679197265242229897
c3: -4523893762164851369
c4: 26886152144094
c5: 594720
Y0: -244291982689000272679978454659345
Y1: 5542638645019034684579869
# MurphyE (Bf=2.147e+09,Bg=1.074e+09,area=2.631e+15) = 4.728e-08
# skew 10405411.22, size 3.066e-016, alpha -8.100, combined = 5.706e-013 rroots = 5
# skew by cownoise
# f(x) = 594720*x^5+26886152144094*x^4-4523893762164851369*x^3-4036779104679197265242229897*x^2+835118785426414901889382598987474*x+3661678253336036003862651470385669302928
# g(x) = 5542638645019034684579869*x-244291982689000272679978454659345
rlim: 50000000
alim: 33500000
lpbr: 31
lpba: 32
mfbr: 60
mfba: 90
rlambda: 2.4
alambda: 3.6

If 15e yields so high that 275M relations arrives before 45M, drop alim to 33.5M.

Indeed this was the case when I first test sieved this job, so please note that alim used a value of 33.5M above (with adjusted alambda).

MQ       Norm_yield      sec/rel       # Rels
10          8969          0.087
14          9822          0.067          38
20         10116          0.069          60
28         10244          0.749          81
36         10055          0.800          81
43          9391          0.082          68
45          9229          0.083          19
48          8916          0.087          27

14-42MQ gives ~280M raw relations

Will enqueue this job on 15e_small in a day or so.

I also ran test sieving on 14e, with the following parameters and the results.

rlim: 50000000
alim: 50000000
lpbr: 31
lpba: 32
mfbr: 60
mfba: 90
rlambda: 2.4
alambda: 3.55

Results of test seving on the algebraic side:

MQ        Norm_yield      sec/rel       # Rels
10           3827          0.057
14           4107          0.056          16
20           4238          0.058          25
28           4390          0.063          35
36           4496          0.071          36
45           4295          0.075          40
55           4044          0.085          42
65           3745          0.082          39
75           3553          0.085          36
80           3488          0.873          18
85           3478          0.112          17

10-83MQ gives ~295M raw relations

I will also enqueue this job on the 14e siever over the weekend.

ETA: Hoping the above data combined with the actual sieving provides data towards improved C170 parameters for CADO.