Cunningham ECM efforts

[chalsall]

Originally Posted by R.D. Silverman

A fundamental (IMO) question remains. Given that this set of 31 numbers is probably beyond the range of NFS@Home, what can be done (if anything) to promote the use of ECM beyond that of the effort provided by YoYo?

It seems that this would be the only means of furthering work on this set of numbers.

If I may please give you a sincere and serious answer to your question(s). In the hopes I can possibly help...

I'm a low-level geek. I deal with compute, networking and storage.

You and yours work in dimensions I only dream in. I'm serious when I say that.

If you and/or others can give me code to run, I and/or others can /possibly/ assist.

I trust that all above parses cleanly... 9^'

[R.D. Silverman]

Originally Posted by swellman

Just noticed that Greg loaded 2,2390L and 2,2350M into the big queue. ECM has not yet been run on these two numbers but I’ll ask Yoyo to remove them from his system. Let NFS@Home do its thing.

ECM has been run; just not the current run. And these numbers were removed. They are not part of the 'gang of 31'
that we believe are too big for NFS@Home.

[R.D. Silverman]

Originally Posted by chalsall

If I may please give you a sincere and serious answer to your question(s). In the hopes I can possibly help...

I'm a low-level geek. I deal with compute, networking and storage.

You and yours work in dimensions I only dream in. I'm serious when I say that.

If you and/or others can give me code to run, I and/or others can /possibly/ assist.

I trust that all above parses cleanly... 9^'

GMP-ECM is readily available for either windows or Linux. However, for the remaing 'gang of 31' we need a mass effort. I was speculating/asking about how to organize such.

[swellman]

Originally Posted by R.D. Silverman

ECM has been run; just not the current run. And these numbers were removed. They are not part of the 'gang of 31'
that we believe are too big for NFS@Home.

ECM was started a few years ago with 9,900 curves @B1=2.9e9 on the then current list of 43 Cunninghams. I know because I was the one who enqueued them. By coincidence, 31 of these have completed ECM, with 12 more to go. You can follow the progress here.

Neither 2,2350M nor 2,2390L have even started ECM at the t70 level. Since each CN takes 3+ weeks to complete ECM at that level, NFS@Home will finish before Yoyo even starts on 2,2350M. I will therefore request Yoyo remove both numbers from his current work queue.

The ‘Gang of 31’ won’t start at the t75 level until ~late Feb 2023. In fact I haven’t loaded them into queue yet in case a factor is found in the these last twelve, now ten, remaining CNs. Hope springs eternal!

[chalsall]

Originally Posted by swellman

The ‘Gang of 31’ won’t start at the t75 level until ~late Feb 2023. In fact I haven’t loaded them into queue yet in case a factor is found in the these last twelve, now ten, remaining CNs. Hope springs eternal!

Please forgive my complete ignorance in this space, but some quick questions...

1. Is there currently an API available where work units can be fetched and returned?

1.2. Like the Primenet et al API(s) / IPC.

2. What the average, and separately, maximum, likely runtime per job?

3. What is the size of any checkpoint file(s) which are generated and backed-up?

4. Is GPU code available for this work?

I presume everyone understands where I'm coming from here... 9^)

The GPU72 TF'ing is comfortably ahead of the Mersenne wave fronts (finally!).

It might be an interesting exercise to expand the options to this compute space. The Colab instances can be exceedingly ephemeral. On the other hand, there can a lot of them...

And... Please tell me if someone has already done this. I know there have been many experiments with Colab over the years. Difficult to keep up with /everything/...

[swellman]

Originally Posted by chalsall

Please forgive my complete ignorance in this space, but some quick questions...

1. Is there currently an API available where work units can be fetched and returned?

1.2. Like the Primenet et al API(s) / IPC.

2. What the average, and separately, maximum, likely runtime per job?

3. What is the size of any checkpoint file(s) which are generated and backed-up?

4. Is GPU code available for this work?

I presume everyone understands where I'm coming from here... 9^)

The GPU72 TF'ing is comfortably ahead of the Mersenne wave fronts (finally!).

It might be an interesting exercise to expand the options to this compute space. The Colab instances can be exceedingly ephemeral. On the other hand, there can a lot of them...

And... Please tell me if someone has already done this. I know there have been many experiments with Colab over the years. Difficult to keep up with /everything/...

Some general info.

Yoyo is a BOINC project based in Germany. Some general info is here. And graphic progress bar (each tab for a different project).

Some tech info is found here as well.

Queues are found here, again organized by project. Keep in mind that up to B1 of 850M the queue is found in the file “wu_status”. For larger B1 values, look in the p1 file. Please note once a job starts processing ECM its status is changed to “DONE” but it can take days to several weeks to completely finish.

But I am only a gatekeeper, you will need to talk to Yoyo (the admin’s handle) for more. He is a member of MF.org, you can reach him via PM.

[chalsall]

Originally Posted by swellman

But I am only a gatekeeper, you will need to talk to Yoyo (the admin’s handle) for more. He is a member of MF.org, you can reach him via PM.

Thank you for the knowledge. Very helpful.

From my POV, the issues with bringing Colab to any effort is the fact they can last as few as two hours each. So the check-points have to be managed (somehow).

I'll reach out to Yoyo when I have some cycles, and see if there's anything we can do.

[pinhodecarlos]

I can see under my yoyo account an option for ECM-P2 which means at the moment only CN are allocated to have ECM phase 2, memory intensive usage. I'll see if I can ask for help.

[LaurV]

Originally Posted by R.D. Silverman

... what can be done (if anything) to promote the use of ECM...

You mean like "give Ryan some beer"?
Or "catch him in a dark alley".... hihi

[pinhodecarlos]

2,1180+ killed

Edit: my message can be deleted since Greg added below more details. Wow!

https://escatter11.fullerton.edu/nfs/numbers.php

[frmky]

2,1180+ is done. It was our first with 35-bit large primes, and everything went surprisingly smoothly. Approximately 3.8 billion raw relations yielded 2.5 billion unique relations. Filtering produced an 82.2M matrix, which took about 39 hours to solve on four nVidia A100's. Square root took about 3 hours for each dependency, and the factors were found on the 4th dependency.

I've added 2,1091+ to the queue.