stephenbrooks.orgForumMuon1GeneralWhich lattice do we kick out?
Username: Password:
Search site:
Subscribe to thread via RSS
Stephen Brooks
2011-04-07 18:52:04
To get a bit more computing power on the 6Xc2 lattice, I'd like to finish one of the other two lattices.  Of those, 7tc2 has the higher score and has been going for only three months.  10d2 has been going over a year, but it's a more complicated optimisation (10 blocks vs. 7) and still shows signs of improvement, whereas 7tc2 is pretty flat at the moment.  However, 10d2 still has the "2 microsecond" time limit that was fixed in 7tc2, so may not ever reach the best linac if it has a large time-of-flight.  Neither lattice is a record-breaking score right now.

I'm tending towards stopping 7tc2, but I'm not sure.  I have to keep at least one going for the time being, so that pre-v4.45 users have something to do (6Xc2 is 4.45+ only).  So you can "vote" here on which lattice you want to get rid of, though I can't guarantee I'll act on the majority in the end
Zerberus
2011-04-09 21:50:34
I've seen that 2µs problem already.  Would it destroy the simulation if I manually raised the timeout value in that lattice file?

Maybe finish both and re-release the 10d2 one as an A-lattice with the 2µs bug fixed.  You could seed it 1:1 using the non-A results.  Of course that's just a proposal.

I actually like 4.45 - the colors don't go crazy anymore under 7 and they actually make sense now (blue for the low-energy and white for the high-energy ones).
Stephen Brooks
2011-04-09 23:06:03
If you raised the timeout manually you'd get a scientifically-valid result but the checksum would no longer match what the Muon1 stats server is expecting, so your results would be removed for being too correct(!)

Your suggestion makes sense - no-one wants to work on a lattice that's either stopped or is artificially limited by a timeout problem.  I could create a Linac900Ext10td2 (note the added "t" ) that has the time limit raised, for pre-v4.45 users to work on for a few months more.  [It would have become 10Xd2 if I'd made it require v4.45+ too, but then the unupgraded users would have nothing to do].

The colours were *meant* to be blue for low energy etc. all along, but from Vista onwards Microsoft has made a hash of handling fullscreen 256-colour modes.  Now that graphics cards are ubiquitous, they can handle the extra work generated by going up to 16M colours with no detriment to the CPU or RAM usage.

The other problem was that pre-v4.45 versions limited to white too quickly for high energies, so there wasn't any contrast above say 600MeV, that's why I tweaked the colour scale.
Stephen Brooks
2011-04-09 23:46:14
Well no time like the present, I guess.

[edit] I'm also going to seed the new 10td2 with one of the best from 10d2. As it's exactly the same machine but with the time limit raised, there isn't much to be learned from letting it run all the way up from nothing again.
Zerberus
2011-04-10 04:42:22
Just be aware that the High-res scan project actually runs on the old 10d2. If it is kicked out High-res scanners could run into trouble.
yoyo
2011-04-10 15:28:12
I updated the Boinc App also to Muon1 version 4.45.
Stephen Brooks
2011-04-10 17:51:55
Zerberus - I told scan users to run the config_nonet.bat file, so it switches off auto-send of results but also auto-update of lattices.  So under those conditions Muon1 would not delete the old 10d2 lattice.

yoyo - thanks.  I think I e-mailed you about the new version, sorry if I forgot.
Stephen Brooks
2011-04-18 14:10:31
Looks like my seeding of 10td2 (with 10d2's best) has created a dead duck lattice that's stuck in a local maximum.  Might replace it in a bit.
: contact : - - -
E-mail: sbstrudel characterstephenbrooks.orgTwitter: stephenjbrooksMastodon: strudel charactersjbstrudel charactermstdn.io

Site has had 16981413 accesses.