Are you sure of your measurements Klaus? For the first script, I get this below (daily snapshot, -j1). As expected, the contact detection takes less time, but since it is less than 1% of the total time, the simulation time is almost the same (still slightly faster in 3001). For the second script, could you output those numbers too (*)? Or even send me the script?
(*) from yade import timing O.timingEnabled=1 O.run(80000,True) timing.stats() bzr2999 Name Count Time Rel. time ------------------------------------------------------------------------------------------------------- ForceResetter 200000 1071391us 0.72% InsertionSortCollider 1549 653503us 0.44% InteractionLoop 200000 79302843us 53.12% "gravity" 200000 7298638us 4.89% NewtonIntegrator 200000 60962689us 40.84% TOTAL 149289065us 100.00% bzr3001 Name Count Time Rel. time ------------------------------------------------------------------------------------------------------- ForceResetter 200000 1057678us 0.72% InsertionSortCollider 761 236650us 0.16% InteractionLoop 200000 77332028us 52.39% "gravity" 200000 7084577us 4.80% NewtonIntegrator 200000 61898584us 41.93% TOTAL 147609519us 100.00% On 24/01/12 03:27, Klaus Thoeni wrote: > Hi Bruno, > > i just tried your new implementation of the optimized contact detection in > the > trunk version. Have a look at the results: > > script: examples/WireMatPM/wirecontacttest.py > 200000 iterations, 1253 bodies > rev 2999: -j1 2min30sec, -j2 1min41sec, -j3 2min15sec > rev3001:-j1 2min37sec, -j2 1min53sec, -j3 2min15sec > > other script: > 22785 bodies > rev 2999: -j1 1min42sec, -j2 1min20sec, -j3 1min29sec > rev3001:-j1 2min38sec, -j2 2min17sec, -j3 2min44sec > > This looks not good to me. I must say I didn't play with the parameters but I > remember when I was testing with your collide2 branch (rev2925) I got much > better performance. > > Any comments? > > Klaus > > On Tue, 24 Jan 2012 02:39:38 AM Bruno Chareyre wrote: >> Hi all, >> >> The branch with optimized contact detection has been merged with trunk >> in bzr3000. >> According to my last tests and roughly speaking, yade is _3 times >> faster_ after this commit, but the result will be worst or better >> depending on the number of particles, the number of threads, the >> constitutive law used, the velocity of bodies, and a few other things. >> >> If you don't compile yade, you can try the new version by installing or >> updating the yade-daily package. >> A few timings are reported on the wiki at >> https://www.yade-dem.org/wiki/Colliders_performace >> I would be glad to get feedback on the actual speedup in other simulations. >> I can imagine possible slowdown in extreme cases, although it is >> unlikely to happen (if it happens, then tuning the collider's parameters >> should fix the problem). >> If you feel like experimenting, you can play with the three parameters >> of the new collider: verletDist, targetInterv, and minSweepDistFactor. >> Reports can be very usefull as they will help to decide what is best >> best default combination for these parameters. >> >> I hope we can release a stable version on this basis soon, when it is >> confirmed that the new collider is bug-free. >> >> Best regards. >> >> Bruno > _______________________________________________ > Mailing list: https://launchpad.net/~yade-users > Post to : [email protected] > Unsubscribe : https://launchpad.net/~yade-users > More help : https://help.launchpad.net/ListHelp > -- _______________ Bruno Chareyre Associate Professor ENSE³ - Grenoble INP 11, rue des Mathématiques BP 46 38402 St Martin d'Hères, France Tél : +33 4 56 52 86 21 Fax : +33 4 76 82 70 43 ________________
_______________________________________________ Mailing list: https://launchpad.net/~yade-users Post to : [email protected] Unsubscribe : https://launchpad.net/~yade-users More help : https://help.launchpad.net/ListHelp
