stephenbrooks.orgForumMuon1Q&ALonger Process times
Username: Password:
Search site:
Subscribe to thread via RSS
slayre
2003-12-12 23:29:20
I recently upgraded to the lastest version (4.33) and I noticed that it takes a lot longer for my computer to reach 200ns instead of doing 10 or 15 MPts it does like 300 Mpts.  Can anyone explain why????

Thanks
Garrett
[DPC]Stephan202
2003-12-12 23:53:27
Better designs take longer to simulate because they involve more particles - on average.  That's why your MPTS value went up as well as your processing time per result.

---
Dutch Power Cow.
MOOH!
Stephen Brooks
2003-12-13 10:57:39
It's also probably because your client automatically downloaded and merged the sample results from the rest of the network, including some of the best.  So you are simulating much more efficient designs more often now.

HB Pencils, also sold as "Moron's Choice" Graphite Cigars.
Ripper
2004-02-18 12:04:09
Does that mean that by downloading the samples that my yield will be higher?  I turned this feature off and now it does an AUTOGEN (#gen=0) and they only give me about 1-15 mpts.  I would like to turn this off, as it seems that my computer is working a lot harder but providing absolutely small results.  I thought the idea was to provide a large number of muons, not less.

As a result, in the same time frame, others process more mpts, while the only thing that increases for me is the number of results and minute numbers of muons.

As the "best performers" are those who process more muons, it seems that this autogen feature is doing me more harm than good.  I realize that you want the variety to give you answers, but there has to be some means of control.  I have 8 computers working this project and can't keep up with people who only have 3 or 4 computers. 

Since I am on a roll, the other problem is that this software isn't very optimized for P4 3GHz plus machines.  My 2.8 GHz machines produce 3 times the work in the same peiod as my 3.0 GHz and 3.2GHz machines.  Seems that this is a waste of processor, even though I have multiple processes (threads) going.  I tried to do the same thing using 1 thread and it is even slower.

Any help on this or possible resolutions to make things better?  If not, then I need to find another project.

THANKS

Iley A. Pullen

=============================
Solution Submission:

SELECT * FROM people WHERE clue > 0

=============================
Stephen Brooks
2004-02-19 12:00:24
quote:
Originally posted by Ripper:
Does that mean that by downloading the samples that my yield will be higher?
Muon yield and Mpts will be higher per result, but you will generate fewer results.  The Mpts is meant to be roughly proportional to calculation time, so from the Mpts point of view (the primary method of 'scoring' for the users of the project), whether you have it switched on or not doesn't matter.
quote:
I turned this feature off and now it does an AUTOGEN (#gen=0)
These are random results, which it uses to start off the file - it'll do only random ones for the first 10 results, then start evolving from there.  Random designs will, of course, have typically bad transmission figures.  They will also have few Mpts each, but on the other hand will take that much less time to simulate.  Obviously if you switch off the feature to download the best results from the rest of the network, your computers will have to work their way up on their own, but sometimes this produces new breeds.  Either way, Mpts per compute time shouldn't be much different, though it is a little lower for low-scoring designs due to some overhead per timestep of the calculation when it is only sending very few muons through.
quote:
Since I am on a roll, the other problem is that this software isn't very optimized for P4 3GHz plus machines.  My 2.8 GHz machines produce 3 times the work in the same peiod as my 3.0 GHz and 3.2GHz machines.
I don't know any reason why it should do that, unless the 2.8GHz machine is an overclocked Opteron Razz Is this using the same version?  v4.33a is quite a bit faster than v4.33.

HB Pencils, also sold as "Moron's Choice" Graphite Cigars.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io RSS feed

Site has had 25162821 accesses.