How large is your problem?? If libMesh is dominating your solve time... then you're problem probably isn't large enough to be distributed over so many processors.
Derek On Feb 11, 2010, at 2:53 PM, Yujie wrote: > Hi, Libmesh Developers, > > When I run the codes in parallel mode, I found the performance is not always > improved with the increase of the CPU number. I tried to find the bottleneck > using the libmesh performance table. I found the problem is from Mesh > Communication::parallel_sort(). > At the beginning (from 2 to 14 CPUs), the cost time of Mesh Communication > takes a small percentage of the total time (Check the following table). > However, when the CPU number changes larger than 14, the performance of Mesh > Communication becomes worse, especially parallel_sort(). I don't know > whether it is normal. Could you give me some advice? Thanks a lot. > > Although there are 8 CPUs in one node of the cluster, each CPU added is > from different nodes compared with previous one by using "-machinefile " in > MPI when I add the CPU. > > CPUs Mesh Communication > 1 > 0 2 > 2.6 3 > 2.6 4 > 2.7 5 > 2.7 6 > 2.7 8 > 2.9 10 > 2.6 12 > 3.4 14 > 4.2 16 > 7.2 18 > 7.2 20 > 8.7 22 > 9.5 24 > 13.8 26 > 15.3 28 > 19.4 30 > 24.2 32 > 34.6 34 > 32.4 36 > 36.6 38 > 39.1 40 > 46.9 > Regards, > Yujie > ------------------------------------------------------------------------------ > SOLARIS 10 is the OS for Data Centers - provides features such as DTrace, > Predictive Self Healing and Award Winning ZFS. Get Solaris 10 NOW > http://p.sf.net/sfu/solaris-dev2dev > _______________________________________________ > Libmesh-users mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/libmesh-users ------------------------------------------------------------------------------ SOLARIS 10 is the OS for Data Centers - provides features such as DTrace, Predictive Self Healing and Award Winning ZFS. Get Solaris 10 NOW http://p.sf.net/sfu/solaris-dev2dev _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
