stephenbrooks.orgForumMuon1GeneralHow to find % done
Username: Password:
Search site:
Subscribe to thread via RSS
[ARS]odessit
2002-08-08 09:33:30
Hi all.  I will try to use my rusty programing skillz to make some crude version of a program similar to "SETI Spy" basically it will check % done and estimated completion time, maybe later on it will grab stats from the web.  So my question is:
How do you figure out % done from results.dat and results.txt or any other files?
I tried looking at C code, but closest thing that I took was C++ so that did not help much.
Odessit
Pascal
2002-08-08 09:45:59
Every second line starts with the yield in percent.  But there's also a good result viewer in the package..

___________________________
1: Athlon TB-C, 1.2 GC/s, 256 MB DDR-RAM, Erazor x², ADSL-Flatrate, NIC Intel, Win 98 SE Mainboard MSI-6380 Rev.  1
2: Pentium III, 600 MC/s, 256 MB RAM, NIC Intel, Win 98 SE
[ARS]odessit
2002-08-08 11:32:09
As I understand it it is a score, % of muons transfered what I am looking for is % of certain simulation or a work unit done.
Pascal
2002-08-08 11:37:51
No, you can't see that, because the particles are simulated, and if they get out of the linac channel, the yield is computed. 
I only may tell you, that the more time a work units takes, the higher the yield gets.
Stephen also wants to make some benchmarks, but at the moment we have the problem of the big differencies in the results.

___________________________
1: Athlon TB-C, 1.2 GC/s, 256 MB DDR-RAM, Erazor x², ADSL-Flatrate, NIC Intel, Win 98 SE Mainboard MSI-6380 Rev.  1
2: Pentium III, 600 MC/s, 256 MB RAM, NIC Intel, Win 98 SE
thormodr
2002-08-08 11:39:09
It depends what you want...
Stephen has it set up so that your results are sent when 99 or 100 are accumulated.  There isn't a numbering of results in the results.txt file.  So you could set up a search of the file for the number of occurances of a text string that is common for each result.  i.e. 'zsrc' or '[v4.' or whatever.  The other alternative is look in the source code and find out how the automatic sender determines when the file has 99 or 100 results... Just a couple suggestions off the top of my head...
Bluumi [SwissTeam.NET]
2002-08-08 12:46:31
I think what odessit search is some OTHER thing.

He search a way to say ... "NOW you have done 30% of the Unit" ...

If one Unit need 10 min.  that is after 3 min.
In the graphical Client it say how long the way of the simulation is (in a early version i saw it).

The CommandLine-Version perhaps has this counter too...

odessit is it that what you search?
[ARS]odessit
2002-08-08 13:45:23
Yes, this is exactly what I am looking for smile
scottsaxman
2002-08-08 14:04:44
Cool!  Good idea, odessit.  I hope your programming goes well, I'm looking forward to the progress monitor!
[ARS]odessit
2002-08-08 15:23:53
So by the looks, there is no output to file that will let me find % done.
Can somebody (Stephen?) explain how the client works.

Here is what the website had to say about this
quote:

After each run of the simulation, the result can be appended to a text file in the two-line format shown below.

zsrc=584;rotsrc=080;d1l=322;s2f=997;s3f=883;s4f=577;snf=490;alt=415;narrow=658;na1=422;na2=962;na3=474;
2.060580 (20936 particles) [v4.11] {E416DD6E}
The first line contains a list of named parameters that determine the accelerator design within the range being considered in the current optimisation.  These have normalised values from 000 to 999, which can be rescaled onto the feasible range for that design parameter to find the actual value used.  They may also be used to switch between different schemes of designs - for instance, the 'narrow' parameter is used to switch between four different algorithms for varying the solenoid radii along the decay channel.

The second line starts with the 'score' for the accelerator.  From the optimiser's point of view, this just has to be a valid floating-point value with larger values corresponding to better designs (e.g. percentage efficiency).  The rest of the line shows the number of particles used, the version number of the simulation software and a checksum of the beamline design file (a comment may also be added). 



Looks like the length of simulation depends on # of particles and parameters used.  Right?  Or is it depending just on # of particles?  OR just the parameters?

Edit:
I guess Chris' idea would work, but I am not positive if all simulations in that 100Kb file take the same time to complete.  If it is true than the problem is not very hard.  But it looks like the speed of completin of this 100Kb file is not linear (not constant), and parameters with # of partcles will change, which in theory would change the speed of calculations from time to time in a same 100Kb file.

[This message was edited by odessit on 2002-Aug-08 at 22:41.]
[ARS]odessit
2002-08-08 15:34:13
hmm another brain fart regarding benchmarking. 
How about creating a sort of work_to_do.ini file with pre-set parameters and use only this for benchmarking?  but this ofcourse will not do anything for % done part of the program...
Stephen Brooks
2002-08-09 02:22:26
There is no way to determine the percentage of a workunit done exactly, because it depends on how many particles survive to later stages of the simulation, which in turn depends on how good the design is.  And if I knew to begin with how good the design is, I wouldn't need to simulate it.  big grin

Probably the best way of determining how far you are towards a send is by the ratio of the results.txt file size over 100000 bytes, which is the threshold that I have programmed Muon1 to send it at.  It would be difficult to get any other data out of the program because results.txt is only written to at the end of a given simulation.


"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"
gogomaus
2002-08-09 02:27:23
In fact time for each single run cannot be determined exactly in advance.
It depends on # of particles and the # of timesteps for each of them to be calculated until it crashes or makes his way through the channel/chicane, which is obviously dependent on individual design parameters (see details from Stephen in thread "Anyone got any benchmarking data for v4.2?".
run times vary much stronger than particles # (more or less proportional to muon-yield-%, but that´s of course not available before ending...).
Progress on KB can be watched simply by use of explorer, one run will need some 1.22 KB of disk space...

this message was written parallel to Stepen´s answer... wink
Kosmaj
2002-08-10 21:37:59
Hi, this is my first post.  I found out about the project from Odessit's post at ARS DC Forum.

Odessit,
Instead of %done, since one simulation takes only 10-15 minutes (1GHz P3), can you make a result viewer that will tell how many results are in the result.txt, how many particles, and what's the best result pending.  Similar to what the ECCP client shows.
[ARS]odessit
2002-08-10 21:54:21
I'll try.
This is just a project to get me in a mood for college so no guarantees.  though what you are asking is easy.
GP500
2002-08-11 02:05:57
hello new in this getting in to the mood to.  (no good program skillz ;( )
but is it not possible to get the numbers that are in the graphical version in to a program.

or better like ECCP that the background version has an icon right onder where you can get those info en set the low/normal/high mode and permanent connection (with file size)
you could get all info of how this is done with eccp on ther page http://www.nd.edu/~cmonico/eccp109/ it's open source and any question's i think chris (project leader) will like to help.


BTW: now we are taking about programming is the program already fully tweaked for top-speed and also Good CPU pesific core's (amd & intel ) they can make use of different things 3Dnow or sse-2 and those things

if this is not so, ther can be a big performance growth (eccp became 70% or more faster after tweaking the client)

T E A M: Grutte Pier [Wa Oars] / Fryslân wer mei Kening \
gogomaus
2002-08-11 02:49:20
I welcome everything that helps increasing calc.  time of the core prog, f.e. processor specific optimisation.
All other items I would consider as features, which are nice but not necessary.
I mostly do run the background version on my home PC overnight or during absence, so I don´t need anything more, that will consume calc.  time. 
My personal belief is that Stephen should concentrate his efforts (and time) onto the simulation prog, which is the main issue.
I do enjoy stats updated semiautomatically, but that´s enough.

*** THIS IS MY PERSONAL OPINION , others are free to differ, of course wink
[ARS]odessit
2002-08-11 08:24:50
quote:


BTW: now we are taking about programming is the program already fully tweaked for top-speed and also Good CPU pesific core's (amd & intel ) they can make use of different things 3Dnow or sse-2 and those things

if this is not so, ther can be a big performance growth (eccp became 70% or more faster after tweaking the client)

http://computer.clubs.nl/GP / Fryslân wer mei Kening \


As far as I know the program is not fully tweaked do to a reason that the core is often changing parameters and Stephen trying to create more intellegent algoritm.
Nothing prevents anybody for optimising this proggy though...except programming knowlege.
GP500
2002-08-11 08:34:37
@Gogomaus #yes but is was looking @ the project benefid and mine .
_____________________________________________________________________
#first i came to the conclusion that allmost all extra-background program items i
mentiond are not a no slowdown for the speed.

#only the reading of progres constantly would be some, so that could be better on
command(how is my progress now). 
#the resaon i would like these featers are a second prog.  is messie and bothering.
#and second it would get more people to participade.

#and what also would be good for the project is a setupprogram.
#1# and not every day new clients that automatically downloaded(less stess on dialuppers) it would be better if it is done once in time with lot of improve ments.
offcourse this is not allways the way to go.  (when thereis a rotten bug)

how large and what core's are downloaded for new version.



PS: the texts i write are mostly straight to the point , can't help that, but don't come back @ that 2 me it's just my stile. 
i don't mean it so though smile smile smile wink wink

pwfeh what a text.
all these things are a big load if only one man must do it, so whe should share that.

T E A M: Grutte Pier [Wa Oars] / Fryslân wer mei Kening \
Stephen Brooks
2002-08-11 15:36:37
Normally I try to do a fair amount between releasing versions.  You just happened to arrive when I had messed up a bit big grin


"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"
TheFinalLoser
2002-08-14 02:57:50
maybe you could use the Nano seconds which are counted...
or the way the particles have moved
cause after 200 and some Nano seconds the simulation is stoped...
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.org
Yahoo: scrutney_mallard
Jabber: stephenbrooksstrudel characterjabber.org
Twitter: stephenjbrooks

Site has had 15715951 accesses.