OFFICIAL "SERVER PROBLEMS" THREAD

[chalsall]

Originally Posted by Madpoo

I'll use M96312371 as an example. When you got the assignment, it didn't have any P-1 done on it yet.

Just putting this out there for consideration...

If people use the usual assignment procedures, no LL'ing assignment will be given which has not been P-1'ed, and probably optimally TF'ed.

This candidate seems to be a bit unusual. ramgeis first did the LL test, then did a P-1 test, and then did a second LL test.

No disrespect intended, but this candidate will need another test by an independent player before the results can be trusted.

[GP2]

Originally Posted by chalsall

This candidate seems to be a bit unusual. ramgeis first did the LL test, then did a P-1 test, and then did a second LL test.

No disrespect intended, but this candidate will need another test by an independent player before the results can be trusted.

ramgeis and Madpoo have been discussing this very exponent and a couple of others, a few messages upstream in this same thread.

[Madpoo]

Originally Posted by chalsall

Just putting this out there for consideration...

If people use the usual assignment procedures, no LL'ing assignment will be given which has not been P-1'ed, and probably optimally TF'ed.

This candidate seems to be a bit unusual. ramgeis first did the LL test, then did a P-1 test, and then did a second LL test.

No disrespect intended, but this candidate will need another test by an independent player before the results can be trusted.

I didn't see an LL test on that exponent, just the two PRP tests.

Anyway, I looked and there are currently 481 exponents available for P-1 work that currently have at least one LL or PRP test completed.

That's weird, and that seems to be the crux of this problem... LL/PRP being done without first finishing the HIGHLY recommended P-1 test to start.

PRP-1 functionality is cool, but if the client doesn't support that or we need to do a double-check of an LL test, then we gotta get those P-1 tests done first on the off-chance it finds a factor.

459 of those 481 are for LL tests, with the remainder having a PRP test done previously. And of course a good amount of them are in that zone of 90M-100M exponents, right where ramgeis' work has been lately and when the issue came up.

I was pointing out that in the case of Prime95 at least, I'm fairly certain when it creates the worktodo entry, if no P-1 has been done yet, the worktodo entry ends with a ",0". Otherwise a ",1" means P-1 is done and isn't necessary. Prime95 will see that and do the P-1.

My guess is those 481 oddballs are from non-Prime95 clients doing work without doing the P-1 first, and if I'm following the check-in code correctly, it sees that and then tries to assign them out *specifically* as P-1 work to make sure it's done first.

It's that type of work that ramgeis was picking up on his next "get assignment" pass, which makes me wonder if that work type is technically being treated the same as a first-time check, rather than a double-check in certain cases.

We could just be at a point where we need to pick up the pace on P-1, or modify the assignment handling on the back end to better cope. I'm glancing at the "get assignment" code and I could be wrong, but the P-1 assignment types seem to be treated like a first-time check, which would normally be true, but isn't in 481 specific cases at the moment.

[ric]

Last night, one of my boxes returned a result for M82911853. Something must have gone wild in the process, since:

• exponent has disappeared from assignments list, as it should
• however, it does not appear neither in "account results", nor in "recent cleared results"
• if I look at exponent details, test completion is reported, but it does not appear in its "history" section.

My local log reports

[Sun Feb 03 00:23:53 2019 - ver 29.4]
Sending result to server: UID: RU/Xeon12C, M82911853 is not prime. Res64: <>. Wg5: 8277296A,29364637,00000000, AID: <>

CURL library error: Operation timed out after 180 seconds with 0 bytes received
[Sun Feb 03 02:48:54 2019 - ver 29.4]
Updating computer information on the server
Sending result to server: UID: RU/Xeon12C, M82911853 is not prime. Res64: <>. Wg5: 8277296A,29364637,00000000, AID: <>

PrimeNet error 40: No assignment
This computer has already sent in this LL result for M82911853

Again, a very minor annoyance, reported just to leave track of it.

[Madpoo]

Originally Posted by ric

...
CURL library error: Operation timed out after 180 seconds with 0 bytes received
...
Again, a very minor annoyance, reported just to leave track of it.

Odd. It seems like the server had a hiccup of some sort writing the raw results line to the table where those live. The result itself was saved (there's a different, more concise table just for the LL results) so it's only affecting the history section which is also where the recent results are pulled from.

I'll have to look at the server logs around that time and see what happened, but judging from the curl timeout, I'd guess there was a SQL blocking issue. If I have to I can manually add that result line into the history.

EDIT: I didn't find anything unusual in the server logs apart from the fact that it logged your result actually coming in 6 minutes later than I expected. So that may have been a network issue between client/server of some kind. Anyway, I manually added the history result.

[xx005fs]

I recently realized that my 90 day average and past 24 hour of GHz-Days submitted was extremely high even though I only submitted a fraction (about 1/2) of the calculated amount. Is that a mistake on the calculation on the server or is it just limited to my computer? Just FYI I submitted about 3000GHz-Days of work today.

[TheJudger]

xxx005fs: doing some work on GPUs an submitting manually? If so than this is normal!

Oliver

[xx005fs]

Originally Posted by TheJudger

xxx005fs: doing some work on GPUs an submitting manually? If so than this is normal!

Oliver

Indeed I'm submitting work on GPU. But it used to be okay like 2 weeks ago. The calculations were normal. BTW how can you submit GPU work without doing manual assignment or manually assign an assignment to a specific worker?

[Mark Rose]

The server knows how much throughput your CPUs should give, based on the hours per day they're configured to run and so on. When you submit GPU work the server isn't aware of your GPUs and so the calculation is off.

[xx005fs]

Originally Posted by Mark Rose

The server knows how much throughput your CPUs should give, based on the hours per day they're configured to run and so on. When you submit GPU work the server isn't aware of your GPUs and so the calculation is off.

Ahhhhh i see now. I always thought they were just the pure GHz-Days numbers. Thanks. I am looking for a way to manually assign things to my GPU as if I would on a CPU (aka giving it a "computer name" and get and receive work with assignment GUID). Is there any way to do that?

[chalsall]

Originally Posted by xx005fs

Thanks. I am looking for a way to manually assign things to my GPU as if I would on a CPU (aka giving it a "computer name" and get and receive work with assignment GUID). Is there any way to do that?

Short answer: No. Longer answer:

The closest thing available at the moment is the MISFIT Application for Winblows, or the mfloop script for Linux. These are quite reliable; many people use them in a "fire and forget" scenario.

Importantly, these won't use a "GUID", but if you only have one GPU this shouldn't be a huge problem. They will be provided with an "AID" (Assignment ID), if you fetch from Primenet. These will also be tied to your particular Primenet and/or GPU72 Account (whichever you choose to use), and so you'll see them on your assignments page, and receive the credit upon completion.

Hope that helps.