mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Data > Marin's Mersenne-aries

Closed Thread
 
Thread Tools
Old 2015-07-23, 05:15   #12
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

29·113 Posts
Default

It also occurs to me that we can help resolve the bad/good ratio by looking at the "suspect" results returned by these systems that are now in need of triple-checking.

First off, it's fun to test these single-checked exponents and find out they were (probably) wrong. That means you might just find a hidden prime. So I get that, and that's one part of this.

Second though is to check the ones those machines did so poorly that the error code was bad enough for the server to mark it suspect and hand it out as a "first time check" again. In those cases we can probably guess that the suspect result is bad, when the 2nd residue didn't match, but that's not always the case, odd as it sounds.

If we go after those, we're (probably) adding more to the "bad" tally of these computers, which helps identify the ones that are doing lousy.

In fact, I had in mind a project I was going to tackle on my own of going after all the 34M exponents that currently need triple checks. There are 281 unassigned exponents needing TCs. Of those, ~ 182 were done by a machine with at least one other known bad residue.

Either taking those 182, or the full set of 281, 34M-35M exponents that need TC's... well, we'd be doing the needed checks on them anyway, and building up the statistical set at the same time.

I *could* do it myself given enough time, but I would love having help. My systems are doing a mix of some of these targeted checks, but I'm also going through *all* of the exponents where Curtis did both a 1st and 2nd check but mismatched. I figured I'd help his team out by figuring out which one was good/bad. That'll take me another couple weeks to complete, I think (there were quite a few).

Sound interesting?
Madpoo is offline  
Old 2015-07-23, 05:19   #13
frmky
 
frmky's Avatar
 
Jul 2003
So Cal

22·33·19 Posts
Default

I took the first 8:

Test=34654643,71,1
Test=34665307,71,1
Test=34665871,71,1
Test=34671661,71,1
Test=34682537,71,1
Test=34683613,71,1
Test=34686947,71,1
Test=34691149,71,1

I'll also switch my ~1000 GHz-day/day of TF over to this once my current assignment expires.
frmky is offline  
Old 2015-07-23, 05:21   #14
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

63158 Posts
Default

Quote:
Originally Posted by Madpoo View Post
Of those, ~ 182 were done by a machine with at least one other known bad residue.
Oh, and along the way we may find a few that wind up needing quadruple checks. I've hit a few of those myself. I just noticed that the "182" figure is inflated a little because some of the exponents are showing up twice... both checks done on them were by somewhat unreliable machines, to some degree or another.

For instance, M34670303, M34757869, M34764817 etc.

These are great because we know one way or another we're increasing the "good" count of one and the "bad" of another, which help both systems "reliability scores".
Madpoo is offline  
Old 2015-07-23, 05:48   #15
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

CCD16 Posts
Default

Quote:
Originally Posted by Madpoo View Post
If we go after those, we're (probably) adding more to the "bad" tally of these computers, which helps identify the ones that are doing lousy.
By way of example, consider:
M34919821

The "suspect" result was returned by a machine that has done 15 bad, 12 good, 11 unknown, and 24 suspect results.

15 to 12 for the bad/good ratio isn't great, but since I was initially targeting higher ratios than that, I wouldn't focus on this right away. Resolving a few more of those unknown or suspect results would help one way or another, to either exclude them or pull them into the query. 4 of those unknown/suspect are under 35M.

Assuming the < 35M stuff goes well and people enjoy doing that kind of triple-checking work, it could go on to higher ranges too. There's no shortage of exponents below 58M needing triple-checks where an "iffy" machine provided one of the results. 1049, give or take (based on a metric of the current "bad > good*0.5" for the systems).

I mean, you get some real doozies like:
M44880113

No disrespect at all towards ArtfulDodger, but his systems are some that I've been deliberately doing triple-checks on already for a while. I id'd those machines months ago as some of the more prolific contributors of bad residues. The system in that case has 110 bad results, 17 good ones, 3 unknown, and 20 suspect. That's just not a good track record, and it's done so much work that it seems really weird to me for some reason.

My analogy I just came up with (so it won't be that great) is like a window washer who spent all this time cleaning a high rise, only to finish and realize he'd been using oil instead of soap and water.

In fact that might be a good case study to look at the distribution of good/bad over time to see if there's any sort of pattern. But I digress. It's late... my wind tends to wander.

Or we might find ones like:
M41732771

Here's a case of a machine with just 1 known bad result, zero good or unknown, but 15 marked "suspect" because of some error code or another. Hard to say how good/bad it is until we triple-check them... but in these cases we know that the double-checks on each of those suspect results didn't match, so I have a hunch all 15 will be bad.

Last fiddled with by Madpoo on 2015-07-23 at 05:51
Madpoo is offline  
Old 2015-07-23, 12:14   #16
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2×4,663 Posts
Default

Quote:
Originally Posted by Madpoo View Post
Or we might find ones like:
M41732771
And I notice that these weren't in your original list; I've just had Spidy bring them in.

Please feel free to post (or PM) me additional lists as you generate them. Takes me almost no time to import the candidates for distribution, and we'd might as well get them done now so the DC/TC'ing can start in earnest.

Edit: Oh, and don't worry about duplication -- Spidy is smart enough to know it already holds a candidate, and will not assign for TF'ing anything still owned by another worker assigned through Primenet.

Last fiddled with by chalsall on 2015-07-23 at 13:10 Reason: s/this wasn't/these weren't/
chalsall is offline  
Old 2015-07-23, 15:36   #17
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

1100110011012 Posts
Default

Quote:
Originally Posted by chalsall View Post
And I notice that these weren't in your original list; I've just had Spidy bring them in.

Please feel free to post (or PM) me additional lists as you generate them. Takes me almost no time to import the candidates for distribution, and we'd might as well get them done now so the DC/TC'ing can start in earnest.

Edit: Oh, and don't worry about duplication -- Spidy is smart enough to know it already holds a candidate, and will not assign for TF'ing anything still owned by another worker assigned through Primenet.
Cool (about the dupe checking on your side).

Yeah, the original exponents I listed had only been checked once. The examples I just mentioned had been checked twice, but both times by suspect machines.

What I probably should do is just generate a list of all exponents and their current TF level where they've had 2 checks already with no matches (under 58M). I imagine in those cases there are going to be some that could use an extra bit or two of TF before someone does the triple check.

And by that I mean *all* those needing triple-checks, not just the exponents where one or two of the tests were done by historically flaky systems.

Going by "somewhat flaky" as a metric, there are around 1,050 of them, but there are currently a total 5,428 unassigned exponents below 58M that have been checked twice with no match.

Which list would you prefer? I'm attaching the full list... I sorted by the current TF bit level and I imagine you'd only have to test half of them to an extra bit or two?
Attached Files
File Type: txt Exponents_needing_triple_checks_2015-07-23.txt (68.9 KB, 260 views)
Madpoo is offline  
Old 2015-07-23, 15:39   #18
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

29·113 Posts
Default

In case anyone was confused about the need to also strategically check exponents that have been tested twice with no match...

I checked in 5 exponents this morning... 1 match and it's now verified. 1 mismatch of an exponent already tested once.

And then these 3 beauties that didn't match the first 2 runs (by the same user for first/second check):
M55570813
M55864219
M56302951
Madpoo is offline  
Old 2015-07-23, 16:07   #19
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

932610 Posts
Default

Quote:
Originally Posted by Madpoo View Post
Which list would you prefer? I'm attaching the full list... I sorted by the current TF bit level and I imagine you'd only have to test half of them to an extra bit or two?
OK, I'm bringing in everything now which isn't already "owned", and which need additional TF'ing.

Might as well get these "off our books", and people seem to really enjoy these little specialized sub-projects.
chalsall is offline  
Old 2015-07-23, 18:50   #20
petrw1
1976 Toyota Corona years forever!
 
petrw1's Avatar
 
"Wayne"
Nov 2006
Saskatchewan, Canada

22·1,117 Posts
Default 1 of my own that did not make the list....

http://www.mersenne.org/report_expon...5555123&full=1
petrw1 is offline  
Old 2015-07-23, 22:04   #21
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

63158 Posts
Default

Quote:
Originally Posted by petrw1 View Post
Yeah, it's assigned to someone so I didn't include it. Through all of this I'm going to try and avoid the assigned ones.

I'll check periodically and see if assignments expire so we can pick them up and assign them to our task force (or whatever we'd call ourselves).
Madpoo is offline  
Old 2015-07-23, 22:39   #22
frmky
 
frmky's Avatar
 
Jul 2003
So Cal

22·33·19 Posts
Default

I just submitted the first two results of the eight I reserved. Both matched the previous runs.

http://www.mersenne.org/report_expon...4654643&full=1
http://www.mersenne.org/report_expon...4682537&full=1
frmky is offline  
Closed Thread

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Double-Double Arithmetic Mysticial Software 50 2017-10-30 19:16
Clicking an exponent leads to 404 page marigonzes Information & Answers 2 2017-02-14 16:56
x.265 half the size, double the computation; so if you double again? 1/4th? jasong jasong 7 2015-08-17 10:56
What about double-checking TF/P-1? 137ben PrimeNet 6 2012-03-13 04:01
Double the area, Double the volume. Uncwilly Puzzles 8 2006-07-03 16:02

All times are UTC. The time now is 10:54.

Sun Nov 29 10:54:31 UTC 2020 up 80 days, 8:05, 3 users, load averages: 1.15, 0.98, 1.00

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.