mersenneforum.org  

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

Reply
 
Thread Tools
Old 2014-10-11, 02:45   #1
Syntony
 
Syntony's Avatar
 
"Tony"
Sep 2014
London, UK

5×17 Posts
Exclamation Primenet doesn't believe that "I fixed the hardware..."

I've been running three cores of an AMD Phenom(tm) II X4 965 Processor on LL Doublechecks and got occasional glitches in hot weather which were eventually traced to a badly fixed North Bridge heatsink. At least one suspect result must have been passed back to the server, because the 'Reliability, Confidence' got reported as 0.98, 1.0

Having fixed the heatsink problem and stepped up the cooling the machine is now 100% reliable, but the 'Reset, I fixed the hardware' option is only effective until the next (correct) result is submitted, at which point the 'Reliability, Confidence' reverts back from 1.00 0.00 to 0.98 1.00. This reversion then means that the machine is never permitted to allocate the smallest doublechecks, a little disappointing as it is processing them at the rate of one per week.

Incidentally, I never succeeded in identifying any incorrect results returned, as all results either matched the first time LL test or were later verified as correct, though as soon as I became aware of the glitches I reran any tests that glitched, just to be sure.

Will this machine ever be accepted as 'Reliable'?
Syntony is offline   Reply With Quote
Old 2014-10-11, 10:59   #2
Zr40
 
Dec 2013

10012 Posts
Default

You get a reliability of 0.98 for submitting an unverified LL test. 0.98 is enough to get assigned the smallest exponents, you only need 0.95 for that (or 0.90 for first-time LL).

Confidence does need to be at least 2.0, and the computer must have returned at least (2 * number of worker threads) results in the last 120 days to be assigned the smallest exponents.

Quote:
Actually, reliability is a rolling average. You get 1.0 for a successful double-check. You get a 0.0 for a proven bad result (should only happen if someone verified the exponent before you submitted your result). You get 0.5 (I think) for an error plagued unverified test. You get 0.98 for a unverified LL test.
(source: http://www.mersenneforum.org/showpos...77&postcount=4)

Last fiddled with by Zr40 on 2014-10-11 at 11:01
Zr40 is offline   Reply With Quote
Old 2014-10-12, 00:04   #3
Syntony
 
Syntony's Avatar
 
"Tony"
Sep 2014
London, UK

5×17 Posts
Default

Quote:
Originally Posted by Zr40 View Post
You get a reliability of 0.98 for submitting an unverified LL test. 0.98 is enough to get assigned the smallest exponents, you only need 0.95 for that (or 0.90 for first-time LL).

Confidence does need to be at least 2.0, and the computer must have returned at least (2 * number of worker threads) results in the last 120 days to be assigned the smallest exponents.


(source: http://www.mersenneforum.org/showpos...77&postcount=4)
Yes, I'd read the post referred to and that was one of the reasons I raised this, as the actual behaviour doesn't seem to match up.

Example: Another of the team machines is also doing LL Doublechecks, and until earlier this week had 'Reliability, Confidence' set as 1.00 0.0, having returned a fair number of matching residues with no errors. It's just returned a non-matching residue (probably because the original LL test returned an erroneous residue) and the 'Reliability, Confidence' has changed to 0.98 1.0 - i.e. the confidence has gone up because a residue didn't match - weird, no? It also seems that the Primenet Assignment Rules page has changed recently, as it used to say that Confidence must be less than 2.0, not more than 2.0. Something seems inconsistent here, the implication is that I have to return two non-matching residues in quick succession before that Confidence becomes high enough to be assigned the smaller tests...
Syntony is offline   Reply With Quote
Old 2014-10-12, 11:00   #4
Syntony
 
Syntony's Avatar
 
"Tony"
Sep 2014
London, UK

5×17 Posts
Question

Had another thought that maybe I'm being a bit of a dumbo here - I just have to wait patiently until enough of whatever events increment Confidence happen so that it increases to reach 2.0. Using the 'Reset, I fixed the hardware' option is probably a bad idea, as it resets Confidence to 0.0 - perhaps it's now become redundant with the modified processor rating scheme?

I guess I might have been mislead by the documentation error that used to mention a requirement of 'Confidence less than 2.0'...
Syntony is offline   Reply With Quote
Old 2014-10-12, 11:28   #5
Zr40
 
Dec 2013

32 Posts
Default

I don't have a source, but I believe confidence refers to the number of results submitted for that computer since last reset, so it is incremented when that computer submits a result.

It's not entirely redundant; after resetting the values, you still need to submit two results even if you already meet the "n results in the last x days" requirement.

Last fiddled with by Zr40 on 2014-10-12 at 11:32
Zr40 is offline   Reply With Quote
Old 2014-10-14, 01:47   #6
Syntony
 
Syntony's Avatar
 
"Tony"
Sep 2014
London, UK

5×17 Posts
Default

Quote:
Originally Posted by Zr40 View Post
I don't have a source, but I believe confidence refers to the number of results submitted for that computer since last reset, so it is incremented when that computer submits a result.
Yes, thanks, this is starting to seem very likely - I'll report back when the next result gets delivered (in about a week).
Syntony is offline   Reply With Quote
Old 2014-10-23, 00:23   #7
Syntony
 
Syntony's Avatar
 
"Tony"
Sep 2014
London, UK

5·17 Posts
Thumbs up Solved!

Well, thanks to Zr40, problem solved.

It seems that 'Confidence' does indeed get incremented with each error-free result submission. A further double-check result submitted yesterday resulted in 'Reliability, Confidence' incrementing from 0.98 1.00 to 0.98 2.00, at which point the smallest assignments became assignable (as the other conditions concerning days of work <=10 and minimum 2 results in 90 days were also met).
Syntony is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
[Patch] "Test/Primenet" prompts improvements on console version Explorer09 Software 2 2017-03-09 04:14
mprime (Linux) doesn't do "affinity" correctly... chalsall Software 16 2014-01-03 15:55
Warning of "possible hardware errors" GARYP166 Information & Answers 4 2010-11-17 17:16
what heppened to "PrimeNet Top Producers Awards" nngs PrimeNet 0 2007-04-25 04:29
Primenet individual account report "current iteration" Peter Nelson PrimeNet 7 2005-02-15 00:12

All times are UTC. The time now is 23:52.


Sun Dec 5 23:52:56 UTC 2021 up 135 days, 18:21, 0 users, load averages: 1.40, 1.33, 1.36

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.