stephenbrooks.orgForumMuon1GeneralNew version (v4.43d)
Username: Password:
Search site:
Subscribe to thread via RSS
Stephen Brooks
2006-07-11 15:44:04
I uploaded this one quite quickly as I found since March (or 4.43) the optimisation algorithm hasn't been working as efficiently as it could.  So try this one, hopefully it's stable.  If a few people try it and it works well I'll send out the upgrade notification alert.
Stephen Brooks
2006-07-11 16:36:58
It can be downloaded from the main page.
HaloJones
2006-07-11 22:36:00
Trying it now.
Xanathorn
2006-07-11 22:52:11
Trying it as well, will post my client's findings in a couple of hours .
Herb[Romulus2]
2006-07-12 06:53:12
That sounds promising
Herb[Romulus2]
2006-07-12 08:47:40
On one box (identically configured as an other one) I encounter exeption errors like:
0xc000001d at 0x0042f1c3
or
0xc000001d at 0x00432a3a

The only difference is that Symantec Antivirus is running on that box automatically currently, I can't stop that beast, not enough rights. 
I seem to remember that earlier clients crashed also sometimes when the AV program was jumping in.  This scan will last more than 2 hours, I'll come back and see what happens when it has finished.
tomaz
2006-07-12 08:55:12
I just posted one in bug reports...I got illegal instruction 0xc000001d at 0x004076d0. NOD antivirus is running on the machine. 
Herb[Romulus2]
2006-07-12 09:22:27
The scan is done, but the client doesn't start anymore, I reverted to the C version and it worked. 
Strange thing is, that on my identically configured working box the new version runs just fine, though I've no feeling yet it is doing any better.
Stephen Brooks
2006-07-12 10:24:40
OK, a debugging version is now available that should generate a more informative error.  Probably I've just left some experimental code in that I should have removed.
RGtx
2006-07-12 11:10:34
Using v4.43c I do not believe that I ever had a queued result, started to think it an extraneous function, since upgrading to v4.43d, a steady trickle, (greater thann 1 in 10 results generated), suggests that the optimization is working well.
Herb[Romulus2]
2006-07-12 11:29:01
Sorry, I can't unpack the rar file here, can you provide a zip?
Stephen Brooks
2006-07-12 12:09:35
There's a ZIP file on the main page now.
Herb[Romulus2]
2006-07-12 12:37:53
The debug version creates a blank screen, with Taskmanager I switched to a fault message:
Exception 0xc000001d
Address 0x40b1cf

A second command window shows a message that "Muon is started configuring" then the whole thing vanishes.
[DPC]white_panther
2006-07-12 13:28:06
same here, win2000 pro athlon XP 3200+
Stephen Brooks
2006-07-12 15:46:00
When are these errors occurring - right at the beginning during configuration?

I've started running the commandline version of it here, but no crashes so far.
AMD-USR_JL
2006-07-12 16:10:51
It crashes on my old P3 on XP SP2 also.  It crashes right when i click on the executable.
AMD-USR_JL
2006-07-12 16:16:39
In the commandline it goes:
"Muon1 started: configuring...
Detected 1 logical processors
Hello,...
loading reults.dat...OK (xxx) results
New Simulation"
Then it dies and this little lcc runtime pops up and says Exception 0xc000001d
Address 0x40b1cf
waffleironhead
2006-07-13 00:01:34
I installed 4.43d to a new directory and ran it instead of the 4.43c setup, to minimise coruption of my past work, anyway, everytime the 4.43d runs its first test it runs it five times(verifing quarantined result) then I get the error

C:\Documents and Settings\Owner\Desktop\dpad1>muon1 -c

Muon1 started: configuring...
Detected 2 logical processors
Hello, [TA]waffleironhead
Loading 'results.dat'... OK (1 results)
New simulation
Searching for auto-saved file...
No file present
[WARNING] Record rejected due to bad checksum: was {A0DFEA403E2D04EEA197B15F}
Continue (Y/N/These/All)?

any thoughts...

if i wipe the results.dat and results.txt the same result happans...I get a bad checksum and dpad stalls
AMD-USR_JL
2006-07-13 02:48:53
I remember i had an error like that before.  I think i had a corrupted .sav file, and all i had to do was delete it.  Or might have just typed A for all.
Herb[Romulus2]
2006-07-13 06:33:41
I have detected a difference in my 2 installations, the client which is working now more than 16 hours is dedicated to decayrotb, while the other one crashing has "none" in the config.txt
Herb[Romulus2]
2006-07-13 09:42:30
I've stopped the running client also, as I found it counts lower than before.  I'd made a number of checks which should have lead to new higher values, but all were in a lower range of -0,004151
to -0,003875 against the previous records, while nearly all earlier steps I checked are normally in a range of -0,000850 to +0,000500
A reverification of 2 leading results prooved it also.

Going back to C now
Xanathorn
2006-07-13 12:37:33
I noticed the lower results also but still giving it a chance.  Didn't get any rechecks with a results.dat with only 4.43c results but I created a results.dat with only 4.43d results and now there are some rechecks again with positive results though the final outcome is still negative.  I'm up to -0.016 now again and hope it to go positive soon and then we'll see how it goes.  By the way I have no problems with the new client (background), running on dualcore AMD on windows XP SP2.

The rejected checksums may be due to a corrupt .sav file or there's an empty line in your results.dat which will also cause that behaviour.
Stephen Brooks
2006-07-13 12:57:47
There will always be some random variation (at least in the PhaseRotD lattice), so small differences aren't really a problem.  I'm more concerned about client crashes.
Stephen Brooks
2006-07-13 13:02:07
I've changed the compiler switch - removed one "optimisation" option (Pentium Pro instructions) that had caused problems in the past.  The resulting beta is now on the main page again.
tomaz
2006-07-13 13:22:01
It is runing now !
Herb[Romulus2]
2006-07-13 14:24:18
Yup, continued without problem on the so far crashing box.

The box is fed with 19 PhaseRotD tests in the queue, so I'll see tomorrow how it compares to previous results.
Stephen Brooks
2006-07-13 14:59:16
Good, OK, I'll risk sending out the signal to upgrade for all users then.  Also make some sort of mental note not to turn that switch on again.
RGtx
2006-07-13 15:50:27
The 'original' d version, is working fine on my P4 3.4G, should I upgrade, or "if it ain't broke, don't fix it"?
Stephen Brooks
2006-07-13 15:57:22
I would install the current version just to be safe - only the EXE has changed.
AMD-USR_JL
2006-07-13 15:57:28
Yeah, should we switch?
AMD-USR_JL
2006-07-13 15:58:18
Sorry, posted too late
Herb[Romulus2]
2006-07-14 06:49:11
The new client definately counts lower on both Decay and Phase. 
PhaseRotD Old client 0,032099 New client -0,052085 Delta -0,084184
I mean that makes the stats irregular when we have C and D version results together.  And you now how small the improvements per step are.
10 out of 19 were better than the recalculated best, so definately small design improvements, but they fall out as exept one all are below zero again.

I reinstalled version C now and do repeat the whole series.  It will lead to some new top results and their combination may produce even a little jump when I'm back on Monday.

tomaz
2006-07-14 08:53:38
RGtx, it would be nice when you can run both and estimate if or how much does Pentium optimization counts...

Regarding versions c and d, I can observe much beter results with ver.  d (phaserotb lattice).  Herb, I think it is a not good test to run same test queues with both versions because you never get same results anyway.  Better way to compare algorithm is to run both versions with same .dat file for a while and see which one progres better...
Herb[Romulus2]
2006-07-14 11:37:49
With version C you can reproduce the queue for Decay perfectly 1:1 and for Phase with some slight differences.  However that is not my point, due to clear different accounting these both versions should not play into one accounting system. 
Given you produce a new good design with version D, it will be nested very much lower in the rankings than in fact a worse design created with version C.
Herb[Romulus2]
2006-07-14 11:53:52
I forgot, a good design on D wouldn't make it anyway in the top list as it might not been rerun anyway.
Stephen Brooks
2006-07-14 13:09:30
I can't find any actual change in the code since v4.43c that would cause different scores, so I'm going to assume for now it's only your setup that is exhibiting this behaviour.
Herb[Romulus2]
2006-07-14 15:05:23
I'll recheck it on my home box over the weekend, which has at least the same processor as these 2 boxes I'm comparing here.
AMD-USR_JL
2006-07-14 15:13:14
So you mean my .32 score on PhaseRotD would have had a higher muon percentage if i had crunched it with the C version?
Herb[Romulus2]
2006-07-15 13:28:23
I can confirm the worse behaviour on a 3rd box now, with version C I've reverified my 0.318605 from yesterday exactly, while on version D it came out now with 0.317600.
For PhaseRotD the picture is more diffuse because the results are not exactly reproducable, but the trend is also clear to see.

I will recheck this behaviour on Monday evening again with 2 complete different setups where I can get my hands on.

: contact : - - -
E-mail: sbstrudel characterstephenbrooks.org
Yahoo: scrutney_mallard
Jabber: stephenbrooksstrudel characterjabber.org
Twitter: stephenjbrooks

Site has had 16395780 accesses.