Index 1 Sequence Work for the "n^i" Aliquot Project

[gd_barnes]

Originally Posted by EdH

For ECM, I just finished 94^100 and there were only 3 of 23 that factored by t40 today.

3 is reasonable.

~10-15% is generally about what I get going from t35 to t40 if I know the range has not already been ECM'd by others. The problem is some of the time, they already had either t40 or full ECM. For example on the same-parities way back when I initialized them, I would ECM to t40 or full if the sequence looked interesting. But most of those remaining here look pretty bad for termination purposes so it's safe to assume they most are likely at t35. As for the opposite-parities, when I did their initialization effort, I did everything to t35. I've completed that for all bases >= 80. Since you're above that now, all opposites should be t35. Of course some of either parity could have been done by others to a higher level at some point.

When I'm doing ECM of random sequences from t35 to t40 where I don't know their ECM status, I might get slightly less than 10% splits.

[gd_barnes]

Sieving to t40 is done for bases 100 thru and including 200.

I'm going to take all bases > 200 to t40. 53 total sequences excluding the C<=135. It should be complete in ~18-20 hours.

Out of 40 sequences taken to t40 today, 7 were split. Not a bad total. But only 1 of 7 initially fully factored; rather unusual. 3 of the 6 partially factored were C<115 and were taken care of quickly. That left 3 remaining:

200^75 (C122) I'm working on.
102^93 (C130) I'm working on.
105^93 (C142) You can have that one. Sieved to t40.

The two above will be done in ~1 hour and I'll report them then.

*****

The C132-C133s...what a fiasco. Check out the aliquiet.exe thread for my problems there. I easily should have had all 8 of them done but I didn't discover the error for ~4 hours and I had to run all 6 C132s with the slower Yafu process. Anyway, 4 of them are done. 4 are still running...2 C132s and 2 C133s. They will be done late morning or so. So I will report them late afternoon.

[EdH]

I haven't run anything since the db locked me out last night. I have taken note of your work on 200^75 and 102^93 - mentioned by me in writing here, so I should keep them in mind (even though they're outside my focus).

Once I get things running again here, I will queue 95^90 through 99^99 for t40 work and then move to the c>134 listing. I should make it past the "105^93 (C142)" that you left for me. 696^56 is within your t40 range, but I see you are excluding it, so it will be included here.

I have been following your exploits in the Aliqueit.exe thread. I haven't posted anything, because I haven't thought of anything constructive to add and Karsten appears to be on top of it better than I would be. But, I will continue to follow the thread and see if I notice anything of interest.

[gd_barnes]

A note: Quite a while back, I fully sieved all C<=135. That is why it is excluded from the t40 sieve. You are good to go on all of the C135s!

200^75 finished 20 mins ago. 102^93 just now finished. :-)

Unfortunately I haven't had time to research the other issue in depth. I have a feeling it's going to involve programming changes to old programs that few people are willing to work with now. That said, surely someone else has encountered a similar problem and maybe has a workaround that does not involve manual intervention each time a C132 pops up. Karsten was able to quickly reproduce it.

[gd_barnes]

All C<134 have been cleared.

I did a quick check of the t40 ECM process. It split one that is right in your wavefront: 220^78. So you can remove it.

It has reached base 780 so everything down to the tail-end of the C145s has been done to t40 so you're in good shape.

I'll be busy for most of the rest of today until late.

[EdH]

Originally Posted by gd_barnes

. . .
I did a quick check of the t40 ECM process. It split one that is right in your wavefront: 220^78. So you can remove it.
. . .

Thanks for letting me know. I don't have index checking in my script (although I thought about adding it), so it would have started working on the c156, unless I noticed.

[gd_barnes]

ECM to t40 is complete for all bases > 100.

Info:

Bases 100-200:
7 out of 40 factored:
102^83
102^93
104^79
105^92
105^93*
120^84
200^75

*All were removed except 105^93 with a C142 cofactor.

Bases > 200:
6 out of 53 factored:
220^78
696^59
966^59
1152^51
14316^39
15015^38

All were removed.

******
Another ECM note: Coming up shortly in your wavefront: 210^66 and 696^54 have both been fully ECM'd. This was done way back in the other same-parity thread. With both of them having starting size and current size of 154 and no factor 3, I listed them as promising. Hence the full ECM.

[EdH]

(Most of) my machines got the night off, since I couldn't get anything from the db up to bedtime. I'll see how they run today. I'll start up the march from the two you mentioned. I'm going to play with t45 on a few of the others, but not sure where I'll head after that. I should be working right in that area for today, though.

I still have some things not quite right machine-wise and I'm putting work into my "BASH polyFind" scripts.

[EdH]

Hmm, 105^93 gave up a 40 digit factor under t45 ECM in about 2 hours. NFS would have taken about 3 hours. But, the ECM work took some of the clients away from the NFS work, so I'd need to study these times more to see if t45 is worth it at this level. Obviously, it saves time if successful, but . . .

[VBCurtis]

Originally Posted by EdH

Hmm, 105^93 gave up a 40 digit factor under t45 ECM in about 2 hours. NFS would have taken about 3 hours. But, the ECM work took some of the clients away from the NFS work, so I'd need to study these times more to see if t45 is worth it at this level. Obviously, it saves time if successful, but . . .

If you're looking into any study of saving time vis-a-vis ECM, I suggest you try:
1. skip the 1e6 curves
2. Do 400 @3e6 to cover the T35 and a bit of T40
3. Do 850 @8e6 for the rest of a T40.

The 3e6 curves are 15-17% of a T40 (depending on which version of ECM used), while 8e6 takes ~1000 curves for a T40 but the 3e6 curves have covered some of the T40. This plan should find more factors above 40 digits, while finding the same number of sub-40 digit factors and with a bit of time savings for the curves.

8e6 is the smallest round number that uses k=2 on the next-biggest chunk of B2. You can get a tiny bit faster T40 with B1 = 7e6 and -k 2 flag to force the larger B2, but it's less than 1% faster and the larger B1 is more efficient at doing a bit of T45 during the T40 effort.

To double-check my work, run ecm -v with B1 = 1e6 and with B1 = 3e6 and compare the predicted time for T35.
Do the same for 3e6 vs 8e6 for T40.

[EdH]

Thanks! I will have to look this over some more. My current ECM effort is divided between a GPU setup and an ecmpi cluster, which has a widely varying number of nodes. The ecmpi is normally driven by Aliqueit values, which I can change, but probably wouldn't at this point.

The GPU has 896 cores (at least that's what GMP-ECM says). I built all my efforts around that number of curves. These are my B1 values for t-levels:

t18 - 1e3
t25 - 15e3
t30 - 12e4
t35 - 1e6
t40 - 6e6
t45 - 5e7
t50 - 4e8
t55 - 2e9
t60 - 2e10

This gives me one GPU pass for each level and the residues are handled by a different machine with lots more threads. While the residues are running elsewhere, the next level is fed to the GPU. I will "play" with those values based on your post and working around the 896 cores in the GPU. I have another (bigger) GPU that is mostly doing LA right now. If I can get ECM working on it, I'll be adjusting based on a greater number of cores when I set that one up.