stephenbrooks.orgForumMuon1Q&AHow does muon1 process work(units)
Username: Password:
Search site:
Subscribe to thread via RSS
GP500
2013-01-09 12:44:29
I can't find a clear explanation of this.

The install is clear: " install, name there, startup-link"

From the read me :
"Submitting Results
The program does this itself!"

But for me it seems like the program never takes on new work (simulation).
And is always doing the same simulation and you can manually input results-samples to get better results.

For me it's like it's just doing the same work(unit) over and over, to get a better result.
And that better result may all ready be done by another.

It's seems to be a wast full method, what am i missing!!!
Stephen Brooks
2013-01-09 16:16:06
The simulations might look similar but they differ in things like the magnetic fields in the solenoids, so the particles follow different paths.

If you look in results.txt or results.dat, these files get longer as new results are generated.  The first (long) line of each result is the parameters and the second line is the score and the "lattice" (whole range of simulations) it is from.
GP500
2013-01-09 18:04:29
So for each simulation on that PC there is a unique configuration?

And how often will this then be repeated (by others), is there a method behind the distribution of simulations.
So that the same work isn't done to often (wasting CPU-time).

PS: looking forward to the 4.46 version.
Zerberus
2013-01-10 04:14:47
The lattices are small text files defining the simulation components and parameters.  Currently there are three different lattices active, however, they may change from time to time (the program downloads them itself).
Though the lattice files define the fixed portion and some limits/ranges, the remaining parameters are varied by the client to get better results.  You could say the simulation does evolute.  To speed up evolution samplefiles with results from other users (kind of a cloud effort) are downloaded, but there may be simulations deliberately omitting samples.
Should your client discover a simulation with a better Mpts result that the highscore, that simulation is queued and repeated five times (by default).  Identical simulations are prevented.
Zerberus
2013-01-10 04:47:24
Correction:

that the highscore -> than the highscore
Stephen Brooks
2013-01-10 10:29:43
More information can be found in Chapter 7 of my thesis
K`Tetch
2013-01-11 03:34:16
And I'll try and write a better, basic overview.  But you can see a list of best designs for the 45-or-so sub-projects we've completed over the last 10 years at <a href="http://muon1.blogspot.com>" target="_blank">http://muon1.blogspot.com</a>

There is a basic overview focusing more on the science of the project in a video of a talk I gave about 18 months ago, at <ahref="http://www.youtube.com/watch?v=5GLk5ph4uBk" target="_blank">http://www.youtube.com/watch?v=5GLk5ph4uBk</a>

But suffice to say, duplicate runs are only done for the 'top' designs, since each sub-project (or 'lattice') has a LOT of possible designs, generally 10^200 or more.  By contrast, RC5-72 which distributed.net has been crunching for almost 10 straight years, at almost 400 billion/second, is at about 2.6%. RC5-72 has about 4 x 10^21

Or to put it again, RC5-72 is 2^72
Linac900Ext6X had 10^198 (1 with 198 zeros) possible designs.  In binary, that's a bit more than 2^650
[Edited by K`Tetch at 2013-01-11 03:42:07]
Zerberus
2013-01-11 08:41:32
Hey K, I think you forgot the HTML button.  Mirroring your post here:
---------------------------------------------------------------------
And I'll try and write a better, basic overview.  But you can see a list of best designs for the 45-or-so sub-projects we've completed over the last 10 years at http://muon1.blogspot.com

There is a basic overview focusing more on the science of the project in a video of a talk I gave about 18 months ago, at http://www.youtube.com/watch?v=5GLk5ph4uBk

But suffice to say, duplicate runs are only done for the 'top' designs, since each sub-project (or 'lattice') has a LOT of possible designs, generally 10^200 or more.  By contrast, RC5-72 which distributed.net has been crunching for almost 10 straight years, at almost 400 billion/second, is at about 2.6%. RC5-72 has about 4 x 10^21

Or to put it again, RC5-72 is 2^72
Linac900Ext6X had 10^198 (1 with 198 zeros) possible designs.  In binary, that's a bit more than 2^650
---------------------------------------------------------------------
[Edited by Zerberus at 2013-01-11 08:42:03]
K`Tetch
2013-01-11 17:03:29
Dammit, you're right, I did.  Not when I posted, but when I edited it (bcause I got the number at the end wrong.

Thanks Z
GP500
2013-01-12 15:22:07
Thx for the insight guys.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 25101371 accesses.