The Next Big Development for GIMPS - Page 22

ixfd64 · 2020-08-02, 04:00

Thanks for the response. I agree that there are plenty of reasons to not deprecate LL assignments for the foreseeable future.

On a related note, my Internet connection isn't the best and randomly drops. If this happens while Prime95 is uploading a proof file, would the upload have to start over? Or is Prime95 smart enough to resume where it left off?

I also realized that I posted my original question in the wrong thread. Sorry about that. It's been moved back here.

Prime95 · 2020-08-02, 04:03

Quote:

Originally Posted by ixfd64

On a related note, my Internet connection isn't the best and randomly drops. If this happens while Prime95 is uploading a proof file, would the upload have to start over? Or is Prime95 smart enough to resume where it left off?.

It will resume where it left off.

Prime95 · 2020-08-02, 04:13

1) Should estimated hours/day in Options/CPU be ditched (moved to undoc.txt)?
Back in the day, it affected what work you might get assigned. It currently affects estimated completion dates in Test/Status. I suspect many users didn't fill the field in anyway.

2) Should the undoc.txt feature "preallocate proof residues file" be moved to Advanced Resources dialog box? It is on by default to make sure prime95 gets the proof disk space it needs.

3) Should the minimum temp-disk-per-worker be 0.0 or 1.0 GB? Currently the minimum is 0.0GB. Should the server use this info in selecting work assignments for the user?
Should a low setting disqualify some work preferences (right now you cannot select 100M tests without setting temp disk to 12.0GB)?

4) I'm fairly certain I want to merge LL-DC and PRP-DC work preference into one DC work preference.

5) Should I add a Advanced/PRP menu choice? Deprecate Advanced/Test (it does LL)? Deprecate any other Advanced choices?

Are there any other obsolete choices or undoc.txt items that should be moved to the menus/dialog boxes? Does anyone use a dial-up modem these days?

ATH · 2020-08-02, 14:01

Maybe leave the older LL tests for LL DC below some limit for example below 65M or 70M then there are plenty of work left for non PRP-VDF users, and then above that people can use PRP VDF if they want.

I do not think it is unreasonable after a grace period to limit first time tests to users with disk space for the proof, it is really a very low requirement, and most computers that cannot live up to this have very old processors anyway.

If Prime95 cannot allocate disk space for proof 8 will it try proof 6 ? I think the option to preallocate space should be left in undoc.txt, it should not be turned off really.

ATH · 2020-08-02, 14:52

Can I run PRP VDF with Gpuowl and turn them in with Manuel Results and then use Prime95 or mprime 30.2 to upload the proof files?

Prime95 · 2020-08-02, 15:55

Quote:

Originally Posted by ATH

Can I run PRP VDF with Gpuowl and turn them in with Manuel Results and then use Prime95 or mprime 30.2 to upload the proof files?

Yes. Or you can use the python script that comes with gpuowl.

kriesel · 2020-08-02, 16:19

Quote:

Originally Posted by Prime95

4) I'm fairly certain I want to merge LL-DC and PRP-DC work preference into one DC work preference.

5) Should I add a Advanced/PRP menu choice? Deprecate Advanced/Test (it does LL)? Deprecate any other Advanced choices?

Are there any other obsolete choices or undoc.txt items that should be moved to the menus/dialog boxes? Does anyone use a dial-up modem these days?

Continue support for features for those who wish to prioritize specifically work on the large LL DC backlog.

Re modems, mere miles from here there are areas with no fiber or cable internet access, leaving dialup (& perhaps slow ADSL), cellphone tethering, satellite or possibly terrestrial wireless mesh as the choices. Some of it is hilly enough to make cellphone service and terrestrial wireless iffy. The Great Lakes and lesser lakes complicate deployment of fiber and cable in the US midwest, among other utilities and infrastructure.

I'd be interested in a specific control to throttle by count or refuse CERT assignments, on a prime95 instance or worker basis, or to specify a minimum exponent for which cert assignments would be accepted, or opting out of PRP-CF certs regardless of exponent. In the past 24 hours I've completed 33 unplanned certs (scattered among 15 systems), and one planned PRP total. Seems more than a little disproportionate. Many of these cert assignments were for tests performed FAR off the wavefront; in the 2.5M to 10.5M range or thereabouts. I don't see much utility in that. Particularly in cases that have already been double-checked and factored, as in https://www.mersenne.org/report_expo...6015563&full=1 The primality status of a lot of these was settled decades ago. Certifying a PRP-cofactor proof does nothing to advance the search for Mersenne primes.

I would throttle my proof work download MB daily quota down, but wonder if that might favor getting more small assignments from far below the wavefront. I suppose I could set daily download to 0 on some systems, except prime95 requires a value between 1 and 100. The cost of a cert itself for these ancient exponents can be trivial at as little as ~0.001GhzD, but the additional Jacobi check or Gerbicz check it triggers in interrupted work on 100M-200M or larger exponents' primality tests is not trivial, especially on older slower cpus.

kriesel · 2020-08-02, 16:25

Quote:

Originally Posted by Prime95

It will resume where it left off.

This is good for colab mprime instances, which terminate unpredictably.

I see 2 of my 3 recently upgraded colab mprime instances have already been commandeered for some CERT crunching.

Prime95 · 2020-08-02, 16:36

CERT crunching can be turned off by setting the download limit to zero MB per day.

Aramis Wyler · 2020-08-02, 16:50

Those are some great questions. I expect that there will be such a large number of computers that will not upgrade prime95 that we should leave all of the lldc to them and eliminate ll first time checks altogether.

You could combine the dc work type, but I would not issue lldc to versions >= 30 any more than I would issue prp dc to versions <30.

Should probably stop issuing ll first time checks altogether, as the work require to DC them is ~= to the work to prp it in the first place; they create almost as much work as they remove, making the contribution almost 0.

firejuggler · 2020-08-02, 21:21

Happy that the cert is all right. But N/A as success? I don't know what should go there ( or if anything should).

2020-08-02, 04:13	#234
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 16557₈ Posts	Random Prime95 changes under consideration 1) Should estimated hours/day in Options/CPU be ditched (moved to undoc.txt)? Back in the day, it affected what work you might get assigned. It currently affects estimated completion dates in Test/Status. I suspect many users didn't fill the field in anyway. 2) Should the undoc.txt feature "preallocate proof residues file" be moved to Advanced Resources dialog box? It is on by default to make sure prime95 gets the proof disk space it needs. 3) Should the minimum temp-disk-per-worker be 0.0 or 1.0 GB? Currently the minimum is 0.0GB. Should the server use this info in selecting work assignments for the user? Should a low setting disqualify some work preferences (right now you cannot select 100M tests without setting temp disk to 12.0GB)? 4) I'm fairly certain I want to merge LL-DC and PRP-DC work preference into one DC work preference. 5) Should I add a Advanced/PRP menu choice? Deprecate Advanced/Test (it does LL)? Deprecate any other Advanced choices? Are there any other obsolete choices or undoc.txt items that should be moved to the menus/dialog boxes? Does anyone use a dial-up modem these days?

2020-08-02, 21:21	#242
firejuggler Apr 2010 Over the rainbow 2·1,303 Posts	Happy that the cert is all right. But N/A as success? I don't know what should go there ( or if anything should). Attached Thumbnails

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Your help wanted - Let's buy GIMPS a KNL development system!	airsquirrels	Hardware	313	2019-10-29 22:51
Is GMP-ECM still under active development?	mathwiz	GMP-ECM	0	2019-05-15 01:06
LLR 3.8.6 Development version	Jean Penné	Software	0	2011-06-16 20:05
LLR 3.8.5 Development version	Jean Penné	Software	6	2011-04-28 06:21
LLR 3.8.4 development version is available!	Jean Penné	Software	4	2010-11-14 17:32

2020-08-02, 04:00	#232
ixfd64 Bemusing Prompter "Danny" Dec 2002 California 5×479 Posts	Thanks for the response. I agree that there are plenty of reasons to not deprecate LL assignments for the foreseeable future. On a related note, my Internet connection isn't the best and randomly drops. If this happens while Prime95 is uploading a proof file, would the upload have to start over? Or is Prime95 smart enough to resume where it left off? I also realized that I posted my original question in the wrong thread. Sorry about that. It's been moved back here.

2020-08-02, 14:01	#235
ATH Einyen Dec 2003 Denmark 2×1,579 Posts	Maybe leave the older LL tests for LL DC below some limit for example below 65M or 70M then there are plenty of work left for non PRP-VDF users, and then above that people can use PRP VDF if they want. I do not think it is unreasonable after a grace period to limit first time tests to users with disk space for the proof, it is really a very low requirement, and most computers that cannot live up to this have very old processors anyway. If Prime95 cannot allocate disk space for proof 8 will it try proof 6 ? I think the option to preallocate space should be left in undoc.txt, it should not be turned off really.

2020-08-02, 14:52	#236
ATH Einyen Dec 2003 Denmark 110001010110₂ Posts	Can I run PRP VDF with Gpuowl and turn them in with Manuel Results and then use Prime95 or mprime 30.2 to upload the proof files?

2020-08-02, 16:36	#240
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 5×11×137 Posts	CERT crunching can be turned off by setting the download limit to zero MB per day.

2020-08-02, 16:50	#241
Aramis Wyler "Bill Staffen" Jan 2013 Pittsburgh, PA, USA 2³·53 Posts	Those are some great questions. I expect that there will be such a large number of computers that will not upgrade prime95 that we should leave all of the lldc to them and eliminate ll first time checks altogether. You could combine the dc work type, but I would not issue lldc to versions >= 30 any more than I would issue prp dc to versions <30. Should probably stop issuing ll first time checks altogether, as the work require to DC them is ~= to the work to prp it in the first place; they create almost as much work as they remove, making the contribution almost 0.