Assign credit to proper computer when using manual submission form

[alpertron]

I'm running P-1 in a computer labeled i5_11400, but Prime95 does not communicate directly with Prime95. I use the manual submission form to send the contents of the file results.json.txt

Each line of this file that is not a timestamp, includes the computer name:

... "user":"alpertron", "computer":"i5_11400"}

But in Account/Team Info > My Account > CPUs these results appear as Manual Testing, not in the row i5_11400.

I think that the Manual Testing row in that report should include only the results where the user does not provide any name for the computer.

[LaurV]

Originally Posted by alpertron

I think that the Manual Testing row in that report should include only the results where the user does not provide any name for the computer.

+1 Same here.

[kriesel]

Same for manually reported GPU results. Even though all reports possible have systemname-gpuid in them. (Gpuowl, mfaktc, mfakto. Some old software doesn't include system info in its report format; CLLucas, CUDALucas, CUDAPm1 for example.) asr2-radeonvii0 is a system-device example. These are included for traceability back to the system and folder they're from. The PrimeNet server lumps all manually reported GPU results into one "manual" bin. Even though that's ~90% of total throughput, from many devices, many software instances, here.

Where's the +100 emoji?

[petrw1]

I have 1 offsite computer for the summer; until September. It has no internet access and I'd like to report results occasionally...and I'd also like them attributed to that PC.

If I copy and submit manually that does not happen as we all know.

So how about copying the prime.spl file to a home PC but before letting it be processed change that PCs name to the remote name....and back after it's submitted.

My theory is they will show up under the desired PC name but a different CPUID. But can I not simply merge them at some time.

[petrw1]

Originally Posted by petrw1

I have 1 offsite computer for the summer; until September. It has no internet access and I'd like to report results occasionally...and I'd also like them attributed to that PC.

If I copy and submit manually that does not happen as we all know.

So how about copying the prime.spl file to a home PC but before letting it be processed change that PCs name to the remote name....and back after it's submitted.

My theory is they will show up under the desired PC name but a different CPUID. But can I not simply merge them at some time.

A flaw in my logic.

I turned off "Use primenet to report..." so I don't get the "Will try again in 70 minutes" over and over.
And now it does not create a .spl file.

[alpertron]

I would like to know if this feature will be implemented.

This year all my contributions were done with the PC named i5_11400, but I still see Manual testing in My account > Results

[Madpoo]

Originally Posted by alpertron

I would like to know if this feature will be implemented.

This year all my contributions were done with the PC named i5_11400, but I still see Manual testing in My account > Results

I'd say it's unlikely, but I'm only speaking for myself. When machines communicate with Primenet automatically, there's a whole deal where new machines will trigger a registration of that new computer, and although there's the friendly computer name that you see, there's a unique guid that belongs to the machine.

The issue I foresee is that a computer submitting results manually could have any arbitrary computer name showing up which may or may not already be registered, coming from any # of actual computers sharing that name. It wouldn't be any better (to me) than just lumping them all into a "manual testing" computer like we currently do because we still have no assurance of what machine the results really came from.

I don't know how many people would actually care about it or not... I know there are some, especially the folks on the forum here who are probably more likely to have multiple computers churning away, some of which are only submitting results manually. But the vast majority of results (for LL and PRP, that is) come in automatically.

The level of effort involved for a small benefit... yeah, not sure about it. At best it could do something like look up the name to see if it already exists (for that user) and if so, use that id, but not go any further with it like create a new cpu ID if it's something new. Meaning the machine name would have had to be created at some point by an instance of Prime95 communicating automatically (there's a lot of stuff it does to generate that unique guid, a "hardware ID" that the client generates, which as some of us have run into, if you change motherboards or something it thinks it's a new machine and creates a new record, etc) so it's not as simple as just popping the name into a table without making even more radical changes.

All of that said, if James or George thought it was a good idea, then more power to them. It would probably be up to them to implement it anyway. So if they wanted to take on that work, good on them. LOL I just know that there are other priorities they'd probably rather spend time on before this bubbled up to the level of "I've got time for this now", but who knows. Like, James and I had recently gone through all of the old results and made sure the application name/version info was being correctly attributed, and now you can see that on the exponent report page. And I recently redesigned the query that page generates so it now loads massively faster than before (and George has been enjoying an awesome winter vacation). That's all on top of the usual little tweaks here and there... making sure TJAOI results are coming in properly (and not being tagged as an ECM or P-1 factor instead of the "TF" factor we use for those results, even though that's still not quite accurate).

[kriesel]

Originally Posted by Madpoo

The issue I foresee is that a computer submitting results manually could have any arbitrary computer name showing up which may or may not already be registered, coming from any # of actual computers sharing that name. It wouldn't be any better (to me) than just lumping them all into a "manual testing" computer like we currently do because we still have no assurance of what machine the results really came from.

Suppose a manually submitted result is determined later to be bad, and the user wants to determine what CPU/GPU it came from, to determine whether the error rate is high warranting intervention, maintenance, decommissioning, whatever.
Last I checked, ~90% of my throughput is GPU. That all gets lumped into a single "manual" bin together, from dozens of GPUs, many of which run from multiple folders and software versions. Even though every manual result submitted where the output format supports it contains a unique name telling what computer, GPU, and even folder it was run on: <computer>-<gpumodel><number>[-w<foldernumber>]. Gpuowl (except for very early versions) supports that included in the result output; mfaktx do; manually performed prime95/mprime would also include that if so configured. Mlucas, ClLucas, CUDAPM1 and CUDALucas don't include that. Mlucas results can be manually edited to add it. Backward traceability would be useful for LLDC and PRP in case of any error.

I don't know how many people would actually care about it or not... I know there are some, especially the folks on the forum here who are probably more likely to have multiple computers churning away, some of which are only submitting results manually. But the vast majority of results (for LL and PRP, that is) come in automatically.

Not from here they don't. A Radeon VII can do 2-3 LLDC or PRP/proof-as-DC per day, and multiple GPUs can be and are installed in one system, while the CPU in the same system takes weeks to do one. All TF here is in mfaktx and manually reported, except for the rare runs for exponents so high they are out of reach of mfaktx, which are way outside of the mersenne.org 1G limit anyway. (100% of TF here is manually reported.)

The level of effort involved for a small benefit...

So saving minutes or hours of sleuthing to find the origins of bad results is considered small benefit? If it's made difficult to track down and resolve error prone hardware, less of that will happen, and that's a loss to the project.

if James or George thought it was a good idea, then more power to them. It would probably be up to them to implement it anyway.

The user time saved going forward might be considerably more than the implementation time. The content is already there, at least in the logs, going back years, if not the database. But yes, their time, their talent, their choices how to spend it.

James and I had recently gone through all of the old results and made sure the application name/version info was being correctly attributed, and now you can see that on the exponent report page.

Yes, something I at least & perhaps others requested, and glad to have it available. Thanks for that. There's been a lot of change in the web pages for user-driven report generation, toward more functionality, and we've noticed.

And I recently redesigned the query that page generates so it now loads massively faster than before

Speed is always welcome (until it costs accuracy.)

That's all on top of the usual little tweaks here and there... making sure TJAOI results are coming in properly (and not being tagged as an ECM or P-1 factor instead of the "TF" factor we use for those results, even though that's still not quite accurate).

Well, you could consider "TJ" for that. Like TF, only different, expressed concisely.

[Andrew Usher]

Yes, I concur. Besides Kriesel's accurate comments above, I would say that 'Manual testing' is, essentially, what gets put in the ID field when there's nothing else to put there. Forcing its use when the CPU/GPU is identified seems not to serve a purpose, though I can see the work that would be involved to change it.

[kriesel]

Originally Posted by Madpoo

The issue I foresee is that a computer submitting results manually could have any arbitrary computer name showing up which may or may not already be registered, coming from any # of actual computers sharing that name. It wouldn't be any better (to me) than just lumping them all into a "manual testing" computer like we currently do because we still have no assurance of what machine the results really came from.

Even with GUIDs and PrimeNet API uc, there is no assurance that computer naming will be unique or stable.
A hundred GIMPS users could name their desktop PC the same. We have separate PrimeNet user IDs.
We can put multiple versions of prime95 on the same system. We can fake or change the system name in the prime95 PrimeNet communications child window. Frequently.
Case in point: the computer now known to the server as "Kilroy_was_here" is in the operating system's point of view, a node named "roa". Which is what the prime95 configuration for it used to include too. Heck, I could change a prime5 system name as often as I like, to "MadpoosMom" etc. (Whatever will fit in 20 characters using the allowed characters. Already changed the test system's primenet name back to "martin" which is a bird species.) I assume someone would intervene if computer names got too inappropriate. I think Madpoo has a sense of humor.

The userID/computername/gpu-name/foldernumber used in gpuowl etc manual reports are unique and stable.