|We're going to do a small experiment with the optimiser. Instead of generating cross-project sample files every 3 hours, meaning we have rapid synchronisation between users and generally only explore one optimisation pathway, I'm changing it to a slower weekly synchronisation. The sample file will be generated on Saturday evening/Sunday morning depending on where in the world you are. For the rest of the week, everyone's Muon1 client is "on their own" to go in different directions.|
|Just out of interest: Does the client recognize if samplefiles are unchanged and won't download them in this case? I changed the check frequency to 168 hours (once a week) nonetheless.|
|It will download them even if they are unchanged, but every 3 hours isn't a lot of bandwidth (the same as the project used before). Setting your client to 168 hours is OK but means you'll download the samplefiles at some random point during week rather than on Sunday like everyone else.|
|OK, I set it back.|
|NOW set it to 168, after it updated today.|
I,m new too this Forum, joined because i have a question !
My english is bad, so reading through the Forum pages trying to find a answer, would be a torture
(but i did almost 2 hours off searching Forum before posting,not finding an answer )
Quest : My client`s use a lot of Waste-time when running, searching for 6X2c_nosample, samplefile ..
http://www.stephenbrooks.org/muon1/samplefiles/linac900ext6xc2_nosample_100.txt . , Txt samplefile ! seems to be empty,, HOW do i make use of the BIN sample file ?????
[Edited by morgan at 2012-10-24 22:49:43]
[Edited by morgan at 2012-10-24 22:59:45]
|linac900ext6xc2_nosample_100.txt is supposed to be empty. Lattices ending in "_nosample" don't use the sample file system.|
|The point, morgan, is to see how the lattice develops with each client working independently, rather than with 90% as a swarm chasing the current high target and the rest giving a small feed to potential new avenues.|
|So we are back to daily? At least the file date on the website updates now updates every day again.|
|It's back to 3-hourly, since I found the _nosample mechanism gave a more distinctive effect on a single lattice. While it was generating weekly, I didn't notice much if any improvement: the results still clustered around the peak, perhaps even evolving a bit slower than before.|
I have to say I'd forgotten about this weekly experiment. Certainly there weren't any stunning improvements in yield as a result.
What the _nosample lattice has done is provided a goldmine of data on how the optimiser gets "stuck" or converges across a fairly big statistical sample of independent runs (one for each user):
...if anything I'd like to get a bit more CPU on this lattice still, since the highest single-users aren't as quite high as our joint 6Xc2 project that went before (3.840515 vs 3.858423). I hesitate to compare with 6tc2 as that was run on pre-v4.45 clients that had some more noise (from a bug) in their output scores.
|Maybe even re-run some of the older 4.44 lattices as _nosample without the noise.|
Anyway, no more open questions.