ECM users - version 30.9-30.13 (see post#316)

[James Heinrich]

Originally Posted by James Heinrich

After quick testing, v30.10b3 Linux64 seems to get CERT assignments again, thank you.

And it has completed several CERT assignments, and just finished the PRP assignment it was working on. But it doesn't seem interested in uploading the associated proof. It's now an hour past completion time, it's checked twice to see if any CERT assignments were available, but the .proof file is still not-uploaded.

edit: as I'm typing this, sure enough it started the proof upload.
But why did it wait 61 minutes between completing the PRP and uploading the proof?
Is this normal and/or desired behaviour?

[masser]

Originally Posted by James Heinrich

Is this normal and/or desired behaviour?

It's normal; I've come to expect a little over an hour for proof file uploads to happen.

[James Heinrich]

I haven't done a lot of PRP, so thank you for the confirmation that it's "normal".
There remains question of whether this is intended behavior or not, and if so, why?

[Prime95]

Originally Posted by James Heinrich

I haven't done a lot of PRP, so thank you for the confirmation that it's "normal".
There remains question of whether this is intended behavior or not, and if so, why?

There is a timer that goes off every hour, a routine is called that looks for *.proof files to upload.
The theory was a timed routine can upload prime95 generated proofs as well as any gpuowl proofs
that are copied into the prime95 directory for uploading.

I might be able to trigger the timer event right after a prime95 proof is generated, or better yet after the
JSON result is sent.

[James Heinrich]

Originally Posted by Prime95

I might be able to trigger the timer event... after the JSON result is sent.

I think if the upload timer could be triggered upon receiving confirmation of successful JSON submission that would be ideal. Still leave the timer to run once an hour but an extra trigger to get the proof uploaded sooner rather than later.

[kriesel]

Why not a simple "upload any available proofs in the working directory now" button?
Advanced, manual communication, []Send proofs now.
That would cover both impatient prime95 users, and gpuowl /prime95 users; drag & drop, send.

[James Heinrich]

Originally Posted by kriesel

Why not a simple "upload any available proofs in the working directory now" button?

That only covers "impatient users" who sit watching Prime95/mprime, it doesn't fix the (real or imagined) problem of the unnecessary 1h delay between PRP completion and proof upload, whereas George's simple proposed solution of triggering the process on JSON upload does solve that.

If George wants to add a menu item in addition that would be fine, useful even, but they're solving two different issues.

[bur]

To prevent mprime from crashing after restarting workers, I now ran 1 worker with 12 threads on a 12 physical cores system. Average CPU load is only about 85-87% on each core. Is there anything that can be done about it?

[kriesel]

Originally Posted by James Heinrich

they're solving two different issues.

On JSON prp/proof result upload, or every (settable in local.txt) x minutes directory check is another way to go. The most impatient gpuowl users could use the manual proof upload web page instead, which is rather immediate.

[Denial140]

I'm not sure if this is the right place to ask/discuss this, but I recently found a factor for M9001 and M8111 with ECM using v30.9, and am interested in running ECM on some other exponents below 10000. The issue is that many of the factors found for exponents in this range were found with significantly larger bounds than the amount of work reported on the GIMPS ECM Progress page, so it is hard to know what bounds to start off with so that I am not wasting calculations - I would guess that many of the other exponents underwent similar efforts without factors found.

The vast majority of these were found by Ryan Propper, and they seem to go up to around M7700. I recall at some point he mentioned on this forum that he occasionally submits results offline to George - I would be grateful if anyone has a rough estimate of how much work this might be.

For example, would I be wasting my time running curves at the 65 digit level (B1=850M) between M1000 and M2000, as the ECM progress page suggests to do? I'm almost certain that the 45 digit level curves (B1=11M) between M4000 and M5000 would be a waste of time, as many of the factors in that range were found with curves at the 60 digit level, but it's hard to know whether I should start at the 65 digit level, for example, or somewhere in between.

Vague rambling over, all advice welcome

[masser]

Originally Posted by Denial140

I'm not sure if this is the right place to ask/discuss this, but I recently found a factor for M9001 and M8111 with ECM using v30.9, and am interested in running ECM on some other exponents below 10000. The issue is that many of the factors found for exponents in this range were found with significantly larger bounds than the amount of work reported on the GIMPS ECM Progress page, so it is hard to know what bounds to start off with so that I am not wasting calculations - I would guess that many of the other exponents underwent similar efforts without factors found.

The vast majority of these were found by Ryan Propper, and they seem to go up to around M7700. I recall at some point he mentioned on this forum that he occasionally submits results offline to George - I would be grateful if anyone has a rough estimate of how much work this might be.

For example, would I be wasting my time running curves at the 65 digit level (B1=850M) between M1000 and M2000, as the ECM progress page suggests to do? I'm almost certain that the 45 digit level curves (B1=11M) between M4000 and M5000 would be a waste of time, as many of the factors in that range were found with curves at the 60 digit level, but it's hard to know whether I should start at the 65 digit level, for example, or somewhere in between.

Vague rambling over, all advice welcome

Exponents between M1000 and M2400 have been heavily worked over by the Cunningham Tables efforts. You might poke around that sub-forum or ask there to get a more complete answer.

For a long time, it was significantly more efficient to run ECM curves for exponents less than about 50000 with gmp-ecm or a hybrid prime95/gmp-ecm. I suspect that a lot of those efforts weren't recorded to the GIMPS database. With the recent upgrades, prime95 is now competitive with gmp-ecm. For very low exponents (like those you're targeting), my general advice would be to run curves at least one, maybe two, levels above what the GIMPS progress report suggests, to compensate for the unreported efforts.

Hopefully others will chime in with additional feedback; I'm curious if there is better advice.