stephenbrooks.orgForumMuon1GeneralIntermittant errors
Username: Password:
Search site:
Subscribe to thread via RSS
Stephen Brooks
2002-10-28 13:08:02
I've just seen a graph some people from Rechenkraft.net plotted of the newest top250 results at http://217.160.138.71/download/muMon/images/top250_plot.gif . It appears something strange is going on: there are _very few_ high-scoring results and the other ones in the top250 form that bunch down at the bottom left.

Huraxprax checked the very highest results (3.586% by astra412) and it turned out to only be 2.9% in the two times he checked it (with 2 different Linux clients).  He also did one by DeltaVee (3.33% originally), and that came out as about 3.3% on one Linux client, but 2.9% on another.

This raises some questions.... It appears that 99.99% of the results I'm getting from you are OK, but there is some sort of intermittent bug happening.  I have fixed several bugs in the current 4.22 source I've got here, and added some validity-checksum.  When it's recompiled, that should reduce the likelihood of this a bit.  I might even restrict the "best muon %" column to only show results from 4.22+ to make sure they are all valid.

If that doesn't fix it, either there is an intermittent bug in the algorithm itself somewhere, or there is some external intermittent effect that is causing the algorithm to malfunction.  I might have to consider getting everyone's Muon1s to re-check their own results too, or at least their highest ones (e.g. if a result is higher than any before in results.txt, re-check it, throw away if new value differs by more than 0.1%, otherwise use average of two values for "true" value).

If you want to use your PC for a project that does not have these problems, you can switch to another DC project for a while and maybe come back when I've got 4.22 or the self-checking version around.
An alternative is to remove the "anomalous" top 15 or 16 entries from the best250 and use that instead.  Hopefully then you will not be stuck trying to repeat astra412's parameters when they seem not to produce high scores.  I.e. if we can be correct in the assumption that the peculiar-looking results are all errors, you can carry on doing correct work using the other results.  I'm not sure to what extent the presence of false high results will hold up the genetic algorithm, but I think it might do.

The best idea might be (if you have a lot of computers) to put all but one of them onto some other project, then when I release new Muon versions use them on the one computer, and if they appear stable maybe transfer some other computers back onto them.

Sorry for the inconvenience - but the project has already been a tremendous success.  Even if 3% is the maximum achievable, that's still a lot higher than the 2.25% we got with version 4.1x, and also a massive improvement on the initial design, which I think scored somewhere in the region on 0.9%!!  We just need to iron out these glitches.  I say 'we' because with 4.22 I will release a utility for you to manually check high results yourself, as an interim measure before the self-checking version.


"As every 11-year-old kid knows, if you concentrate enough Van-der-Graff generators and expensive special effects in one place, you create a spiral space-time whirly thing, AND an interesting plotline"
[ARS]odessit
2002-10-28 15:18:58
frown
I may not understand how the checking works, but here are 5 possibilities.

1. top results by astra412 were manually edited (cheated)
2. Double Check is not working correctly
3. Only Windows client itself has a bug.
4. Only Linux client has a bug
5. Bug is present in both clients (logic error).

I think you should take a look at all possibilities

astra412, I am NOT saying that you are cheating, but just merely listing all possibilities.

ARS Team Atomic Milkshake
Unofficial Muon1 FAQ
Bill[Romulus2]
2002-10-28 16:29:51
It's just the top 16 results that are questionable?  If we edit out those results the rest of us should be OK?
[Ars]SmilingJim
2002-10-28 16:33:06
Why not just release another top 250 with the removal of the questionable entries?
Shades of Grey
2002-10-28 17:55:11
I might've missed what I'm about to say, and if not, well I feel like it should be said.

With the end of Eccp, and rc5-64 you are seeing a lot of new people.  The chances of cheating are much higher.

Just a fyi, I'm sure you've probably thought of this
astra412
2002-10-28 18:55:43
Hi,

Would you recommend removing the new best250 from our .dat then? 

Astra412

(and don't worry about asking about cheating, it's entirely reasonable)
pvs
2002-10-28 19:55:25
0.005638 (32117 particles)
0.599898 (41317 particles)
1.001912 (44946 particles)
1.506862 (44874 particles)
2.051369 (55226 particles)
2.514176 (60277 particles)
3.008696 (59063 particles)
3.586325 (75176 particles)

0.0xx => 0.5xx ~10.000 Particles
0.5xx => 1.0xx ~5.000 Particles
1.0xx => 1.5xx ~0
1.5xx => 2.0xx ~10.000
2.0xx => 2.5xx ~5.000
2.5xx => 3.0xx ~0
3.0xx => 3.5xx ~15.000

75.000 Particles are not really a surprise for a Transferrate of 3.5.
mad-ness
2002-10-28 23:47:55
So, we should pull out the top 16 results ?

Ars Technica
Team Atomic Milkshake
http://www.teamatomicmilkshake.com/index.php
[DPC]Scorpion
2002-10-29 02:39:03
Yeah Please tell us if we need to edit the best 250 en just cut out the top16 results

or that you are gonna release a new top 250 for now

best thing i think is that you release a new version

big grin, also because there are pretty much new best muon's already in the last 2 days

PSEUDO [SwissTeam.Net]
2002-10-29 03:04:05
quote:
Originally posted by pvs:
(List of values cutted)
75.000 Particles are not really a surprise for a Transferrate of 3.5.


Yes sure, for a value of 3.5, might this be possible.
But there are some other results, like this one:
3.047972 (78299 particles)
which is a quite strange result, if you compare it with mine:
3.056100 (59401 particles)


Grüessli
Kay
Stephen Brooks
2002-10-29 04:46:19
quote:
Originally posted by odessit:
2. Double Check is not working correctly


At present, there is no double-check, and there won't be in v4.22. However, 4.22 will have some other bugs fixed that may be causing this current anomaly, eliminating the need for a double-check.  If they don't fix it, I'll need to release a further version incorporating such a check.

--[It's just the top 16 results that are questionable?  If we edit out those results the rest of us should be OK?]--

Yes, that's what I think.

--[Why not just release another top 250 with the removal of the questionable entries?]--

That might be an idea.  OK, I've put the file from rank 17-250 on the server now.

--[With the end of Eccp, and rc5-64 you are seeing a lot of new people.  The chances of cheating are much higher.]--

That's true, although actually I haven't seen any signs of cheating _so far_. I don't personally believe this glitch is due to cheaters.

--[Would you recommend removing the new best250 from our .dat then?]--

Just take out the rank 1-16 entries, or get rid of it and replace by the one I've just put here.

--[So, we should pull out the top 16 results ?]--

Yes.  In a sense I'd like to put this project on "standby", in other words currently the bulk of your cycles would be "better" used on another project, but perhaps keep one machine running on this until I've got 4.22 sorted.
However, doing that (removing top 16) is actually the best way of continuing work in a way that is almost certainly correct.

This probably won't stop those who are only in it for the points, and I see nothing too bad about that (because when I correct these bugs, that large firepower will be directed in the right direction).  Just as long as those people don't suddenly start complaining that they've "wasted CPU time" because of the glitches, because I've now told you they're here.


"As every 11-year-old kid knows, if you concentrate enough Van-der-Graff generators and expensive special effects in one place, you create a spiral space-time whirly thing, AND an interesting plotline"
mackerel
2002-10-29 06:18:20
quote:
Originally posted by odessit:
frown
I may not understand how the checking works, but here are 5 possibilities.

1. top results by astra412 were manually edited (cheated)
2. Double Check is not working correctly
3. Only Windows client itself has a bug.
4. Only Linux client has a bug
5. Bug is present in both clients (logic error).

I think you should take a look at all possibilities

astra412, I am NOT saying that you are cheating, but just merely listing all possibilities.

http://www.teamatomicmilkshake.com
http://www.teamatomicmilkshake.com/faq.htm


Option 6: Hardware error
Two possible causes of hardware related errors which might go unnoticed in normal use are:
(1) components which are damaged or faulty
(2) excessive overclocking

[DPC]Scorpion
2002-10-29 06:22:40
ok Ty stephen

but when i look @ stats you will see this


100. [DPC]Scorpion 0h - - - 361 202`468`844 (.199%) 3.030% (65th)


this means i processed the 65th best result ...but when i look in your newly created top250 i'm not there confused

i know i might be asking to much from you but would you consider to remake the top250 ....because last 2 days some pretty good results came by

I would be very delighted

BrettJB
2002-10-29 08:05:58
Hi Stephen,

As long as we remove the top 16 results (or download the new file), we're no worse off than we were before you published the latest file.  So I'm not sure that switching to other projects is truly necessary at this time. 

Personally, I'm more than willing to keep my machines crunching away, if you think the additional data will help track down the source of the error (if one exists).  I'm also more than happy to beta test any new versions with the clear understanding that those results I produce would not "count", as they may be suspect due to errors.

I'm not really in this for the numbers (though a bit of friendly competition is admittedly fun).  I'm in this because the project fascinates me.

--Brett
AySz88
2002-10-29 09:00:46
quote:
Originally posted by [DPC]Scorpion:
but when i look @ stats you will see this


100. [DPC]Scorpion 0h - - - 361 202`468`844 (.199%) 3.030% (65th)


this means i processed the 65th best result ...but when i look in your newly created top250 i'm not there confused



Not really.  It means you have the 65th best TOP result, not counting the ones you have that are 2nd, 3rd, etc. One person can have more than one result, up to 10 (I think), in the best250. So the results from other people can push yours off the 250.
[DPC]Scorpion
2002-10-29 13:56:57
well look over @ Dukebox great stat

and see this nice list Muon Overall

Bill[Romulus2]
2002-10-29 17:59:51
I'm content with just removing the top 16 results and continuing the project.  Thank you for the reply.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 26761365 accesses.