Thanks for reporting Anton. Actually, you should get a better factor by comparing "old"/4 threads and "new"/4 threads. The strategy of the new colliding mechanism is to maximize the usage of Verlet distance to put more load on the interaction dispatcher (more virtual interactions), and less load on the collider (lower colliding frequency). Since the collider is not parallel but interaction dispatching is, the change is in favor of big parallelized simulations.
It explains the key points of the change: - revisited verlet stride to allow large Verlet dist - optimized handling (loop/dispatch/erase) of virtual interactions Cheers. Bruno On 06/02/12 20:50, Anton Gladky wrote: > Hi, > this is the result of my "working" simulation. > About >100 000 bodies (I guess). > > Thread iter/sec > 1 3,42 > 2 5,61 > 3 7,57 > 4 9,11 > > The old speed on 1 thread is about 1,62 iter/sec > So speedup factor is more than 2! > > The good news, that it is paralleled very good. > The dependency is not linear, but anyway.... > > Thank you, Bruno! > > We just need to test it more precisely, but it seems it is a good major > change. > > Anton > > > > 2012/2/1 Christian Jakob <[email protected]>: >> Zitat von Bruno Chareyre <[email protected]>: >> >>> Thanks! >>> You would actually outrun PFC by a large factor if you could compare >>> with, let's say, 80k spheres. I have no way to test PFC on my side >>> unfortunately. >>> Please, please, try it! :) >> >> Ok, I will try that soon. >> I am thinking about two models for speed test, one quasi-static model and a >> dynamic one. Has anyone an idea how a simple dynamic model for a speed test >> could look like?! >> >> >>> Bruno >>> >>> p.s. Please reply to the mailing list, not to me directly. If you are >>> using thunderbird, there is "reply to list" button, else you have to >>> type the mailing list adress I guess. >> >> Sorry, I was clicking on the wrong button ... >> >> >>> >>> On 01/02/12 14:50, Christian Jakob wrote: >>>> Great work! >>>> >>>> I reactivated my speed test with bzr3011 and got a speed-up between 10 >>>> and 20 %. I updated the comparison page: >>>> >>>> https://yade-dem.org/wiki/Comparisons_with_PFC3D >>>> >>>> The new collider version makes yade up to 20 % faster than pfc! It >>>> seems, that you are outrunning pfc ;) >>>> >>>> Zitat von Bruno Chareyre <[email protected]>: >>> >>> >>> _______________________________________________ >>> Mailing list: https://launchpad.net/~yade-users >>> Post to : [email protected] >>> Unsubscribe : https://launchpad.net/~yade-users >>> More help : https://help.launchpad.net/ListHelp >>> >> >> >> >> _______________________________________________ >> Mailing list: https://launchpad.net/~yade-users >> Post to : [email protected] >> Unsubscribe : https://launchpad.net/~yade-users >> More help : https://help.launchpad.net/ListHelp > _______________________________________________ > 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
