mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > PrimeNet

Reply
 
Thread Tools
Old 2019-01-29, 22:18   #1629
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

2·1,637 Posts
Default

Quote:
Originally Posted by GP2 View Post
How does it automatically assign exponents in such a high range?

Wouldn't it normally give you something in the 83M range?
I still think it's just a case of the result being considered "unassigned result" ergo "poached", so the original assignment is converted into a double-check automatically.

What that tells me is that the code that should look at the assignment ID and match it up is not matching up and accepting it as a completed assignment.

Since this only seems to happen with PRP that's where we'd have to look.

I could, of course, be entirely wrong about my theory. But just turning in a result won't create a brand new assignments... there's nothing in the manual result processing capable of such a feat, aside from the aforementioned conversion of a poached assignment to a DC. I'd have to peek at the assignment IDs and raw results themselves to know for sure, so maybe if I get a bee in my bonnet about it tonight, I'll do just that.
Madpoo is offline   Reply With Quote
Old 2019-01-30, 01:03   #1630
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

2·1,637 Posts
Default

Quote:
Originally Posted by Madpoo View Post
...I could, of course, be entirely wrong about my theory. But just turning in a result won't create a brand new assignments... there's nothing in the manual result processing capable of such a feat, aside from the aforementioned conversion of a poached assignment to a DC. I'd have to peek at the assignment IDs and raw results themselves to know for sure, so maybe if I get a bee in my bonnet about it tonight, I'll do just that.
I'm probably wrong about my theory... I looked at the raw data and it doesn't appear to be a case of an assignment being auto-assigned as a DC because of a "unassigned" result coming in.

The original result comes in and then it's just over a minute before the new assignment is created. That's odd... it's a manual result, not the Prime95 client checking it in, so it's not like the client is auto-reserving the same work again.

I will say this though... when I look at the web logs, I see you check in the result on the manual_result page, and then I see you hit the manual_assignment page and make a request for work with a starting exponent size just smaller than exponent that was just checked in:
Code:
exp_lo=96309630
So M96312371 may very well have been the next exponent available again, and what's likely happening is that it's not being recognized as having a first-time check yet.

There really shouldn't be any time delay in there... a result is committed right away and with a minute+ until the assignment is made, there's no reason it should be assigning the same exponent as a first-time PRP test (and note, I don't think I, or George, added in the same protections on PRP assignments to ensure DCs don't go to the same user who did the first one).

There is a process that happens when a result is checked in where it will take that exponent and do a "make available" on it... if it was a first-time result, it *should* make it available as a DC. The bug (if that's what it is) could be in that process, where it's being made available as another first-time assignment, rather than a DC. That would be my starting point to research next.

For now, I'd say if you want to avoid the problem, when you request a new assignment, set your minimum exponent size to something larger than what you just turned in, and we'll investigate in the meantime.
Madpoo is offline   Reply With Quote
Old 2019-01-30, 16:27   #1631
ramgeis
 
ramgeis's Avatar
 
Apr 2013

11101002 Posts
Default

Quote:
Originally Posted by Madpoo View Post
There really shouldn't be any time delay in there... a result is committed right away and with a minute+ until the assignment is made, there's no reason it should be assigning the same exponent as a first-time PRP test (and note, I don't think I, or George, added in the same protections on PRP assignments to ensure DCs don't go to the same user who did the first one).

There is a process that happens when a result is checked in where it will take that exponent and do a "make available" on it... if it was a first-time result, it *should* make it available as a DC. The bug (if that's what it is) could be in that process, where it's being made available as another first-time assignment, rather than a DC. That would be my starting point to research next.
I noticed something which might be relevant here. All three exponents which show this behavior (96312169, 96312193, 96312371) didn't had a P-1 result report in the first attempt, only the PRP result. I can't tell why this is the case but if this is somewhere in the logic a requirement to set the right flags and stops processing otherwise...
ramgeis is offline   Reply With Quote
Old 2019-01-30, 18:35   #1632
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

2×1,637 Posts
Default

Quote:
Originally Posted by ramgeis View Post
I noticed something which might be relevant here. All three exponents which show this behavior (96312169, 96312193, 96312371) didn't had a P-1 result report in the first attempt, only the PRP result. I can't tell why this is the case but if this is somewhere in the logic a requirement to set the right flags and stops processing otherwise...
That's a good point. When I was looking through the process last night, I did see that when a result come in, the code to make it available again will see that no P-1 work has been done and make it available for that work-type.

How did you get the original assignments? Was it through the manual assignments page? It seems weird that it would have given you those PRP assignments even though no P-1 had been done yet.

Either something went wrong there, or there are exponents out there needing P-1 work that are currently setup to be handed out as LL/PRP.

Consider a sample worktodo entry that looks like:
DoubleCheck=<optional assignment ID>,exponent,<current TF bit level>,<P-1 tested>

The "P-1 tested" area indicates if P-1 has been done at all. If necessary, Prime95 knows to do it before doing the LL/PRP test.

How a GPU client does it is probably way different... I bet if you looked at the worktodo entry you got, it had a zero there and the client *should* have known to do that first since it could find a factor and save the effort of 1 or 2 checks. But that'd be up to the program to know what to do.

So that could very well be a part of it. I guess if you see any PRP assignments come in without P-1, do that first before PRP, or stick to exponents that already had P-1 done.
Madpoo is offline   Reply With Quote
Old 2019-01-30, 18:50   #1633
SELROC
 

7×11×61 Posts
Default

Quote:
Originally Posted by Madpoo View Post
That's a good point. When I was looking through the process last night, I did see that when a result come in, the code to make it available again will see that no P-1 work has been done and make it available for that work-type.

How did you get the original assignments? Was it through the manual assignments page? It seems weird that it would have given you those PRP assignments even though no P-1 had been done yet.

Either something went wrong there, or there are exponents out there needing P-1 work that are currently setup to be handed out as LL/PRP.

Consider a sample worktodo entry that looks like:
DoubleCheck=<optional assignment ID>,exponent,<current TF bit level>,<P-1 tested>

The "P-1 tested" area indicates if P-1 has been done at all. If necessary, Prime95 knows to do it before doing the LL/PRP test.

How a GPU client does it is probably way different... I bet if you looked at the worktodo entry you got, it had a zero there and the client *should* have known to do that first since it could find a factor and save the effort of 1 or 2 checks. But that'd be up to the program to know what to do.

So that could very well be a part of it. I guess if you see any PRP assignments come in without P-1, do that first before PRP, or stick to exponents that already had P-1 done.

This is a good point too. GpuOwl can do P-1 and PRP now, not sure if it does so automatically or if it needs manual intervention to do P-1 before PRP.
  Reply With Quote
Old 2019-01-30, 19:09   #1634
James Heinrich
 
James Heinrich's Avatar
 
"James Heinrich"
May 2004
ex-Northern Ontario

BCD16 Posts
Default

Quote:
Originally Posted by SELROC View Post
This is a good point too. GpuOwl can do P-1 and PRP now, not sure if it does so automatically or if it needs manual intervention to do P-1 before PRP.
Cleverly, it can also do P-1 and PRP simultaneously.
James Heinrich is online now   Reply With Quote
Old 2019-01-30, 22:35   #1635
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

2×1,637 Posts
Default

Quote:
Originally Posted by James Heinrich View Post
Cleverly, it can also do P-1 and PRP simultaneously.
Oh yeah, forgot about the PRP-1, as the cool kids are calling it.
Madpoo is offline   Reply With Quote
Old 2019-01-30, 23:14   #1636
ramgeis
 
ramgeis's Avatar
 
Apr 2013

22×29 Posts
Default

Quote:
Originally Posted by Madpoo View Post
How did you get the original assignments? Was it through the manual assignments page? It seems weird that it would have given you those PRP assignments even though no P-1 had been done yet.
The procedure is always the same. I request some manual assignments for doing first time PRP tests, copy/paste them to the worktodo file of the Prime95 instance (v29.4 build 8) and when they finish I submit them via the manual check-in page. That's it...
ramgeis is offline   Reply With Quote
Old 2019-01-30, 23:24   #1637
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

327410 Posts
Default

Quote:
Originally Posted by ramgeis View Post
The procedure is always the same. I request some manual assignments for doing first time PRP tests, copy/paste them to the worktodo file of the Prime95 instance (v29.4 build 8) and when they finish I submit them via the manual check-in page. That's it...
Okay... I just had to double-check because people can be unpredictable at times, and it's good to check my assumptions now and then.
Madpoo is offline   Reply With Quote
Old 2019-01-31, 06:55   #1638
SELROC
 

3×11×281 Posts
Default

Quote:
Originally Posted by James Heinrich View Post
Cleverly, it can also do P-1 and PRP simultaneously.

I think PRP-1 has been removed in favor of P-1 and PRP3.
  Reply With Quote
Old 2019-02-02, 23:47   #1639
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

1100110010102 Posts
Default

Thinking on it further, I have a theory for ramgeis' problem... I'll have to look closer at various components including how a PRP result is checked in and how it's making that exponent available for the next assignment.

It could be a combo of things.

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

You used Prime95 29.5 build 8 so PRP-1 isn't a part of the equation at all.

So, you get a PRP assignment and I guess unlike an LL test, Prime95 didn't decide to do a P-1 test first. Not sure why.

So you do the PRP test, turn it in, and when making it available for the next person it's set as "P-1 needed" and for whatever reason, you get that same one for P-1 work, which it does, and then starts on a second PRP test.

The results you turned in for the P-1 and 2nd PRP test both have the same assignment ID so that's definitely the case on that round.

Anyway, might have to check with George to see why that first PRP test didn't do the P-1 first like it should have, and only did that the next time around.

I think doing it in that order *should* correct the way it's being assigned out a 2nd time as a "first time check" instead of a double-check, which is totally skewing things.

I'll have to take a look at the "make available" code to see if there's something that can be done to pick up on that odd scenario and deal with it there.
Madpoo is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Official "Faits erronés dans de belles-lettres" thread ewmayer Lounge 39 2015-05-19 01:08
Official "all-Greek-to-me Fiction Literature and Cinema" Thread ewmayer Science & Technology 41 2014-04-16 11:54
Official "Lasciate ogne speranza" whinge-thread cheesehead Soap Box 56 2013-06-29 01:42
Official "Ernst is a deceiving bully and George is a meanie" thread cheesehead Soap Box 61 2013-06-11 04:30
Official "String copy Statement Considered Harmful" thread Dubslow Programming 19 2012-05-31 17:49

All times are UTC. The time now is 18:40.

Tue Sep 22 18:40:13 UTC 2020 up 12 days, 15:51, 1 user, load averages: 2.08, 1.89, 1.94

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.