Hello Alexander, > Am I right that if one wants distributed triangulation, dofhandler and > solver one doesn't need metis at all? Then metis is just a part of the > "local triangulartions + parallel solver" scheme, isn't it?
yes; you can use deal.II + p4est (distributed "mesh oracle") and Trilinos (parallel LA + Solvers). I think step-32 is an example for that case. it is not that easy to compile all of these together, I have some install scripts, if you want them please reply. Best Uwe On Sun, 2012-07-22 at 12:26 +0200, Alexander Grayver wrote: > On 21.07.2012 20:12, Wolfgang Bangerth wrote: > > In general, I think METIS isn't on anyone's "most important" list any > > more since we have the massively parallel framework that builds on > > p4est and doesn't need METIS. We should of course continue to support > > METIS nevertheless. > > Hello Wolfgang, > > Am I right that if one wants distributed triangulation, dofhandler and > solver one doesn't need metis at all? Then metis is just a part of the > "local triangulartions + parallel solver" scheme, isn't it? > > > I suppose it would be ok if we drop support for 4.0 and only support > > 5.x. Alexander, I don't know how familiar you are with autoconf but if > > you'd like to play, do you want to give it a try to modify aclocal.m4 > > so that the paths for METIS all fit those for 5.0? We'd certainly be > > happy to accept patches that go in this direction (even if they're not > > completely finished yet). > > I was able to compile deal.II with metis-5.0.2 using attached path. > However running step-17 there seem to be problems: > > agrayver@agrayver-VirtualBox:~/lib/deal.II-dev/examples/step-17$ > mpirun.mpich2 -n 2 ./step-17 > Cycle 0: > Number of active cells: 64 > Input Error: Incorrect ncuts. > Input Error: Incorrect ncuts. > Number of degrees of freedom: 162 (by partition: 162+0) > Solver converged in 9 iterations. > Cycle 1: > Number of active cells: 124 > Input Error: Incorrect ncuts. > Input Error: Incorrect ncuts. > Number of degrees of freedom: 302 (by partition: 302+0) > Solver converged in 17 iterations. > Cycle 2: > Number of active cells: 238 > Input Error: Incorrect ncuts. > Input Error: Incorrect ncuts. > Number of degrees of freedom: 570 (by partition: 570+0) > Solver converged in 26 iterations. > ... > > Two possibilities: > > 1. Most probably I did something wrong. > 2. In metis-5.x there were some changes in functions interfaces. For > instance, I came across with this issue trying to MUMPS with new metis-5.x > > Any ideas? > > > I'm sure you know that I'll be on visit at the GFZ on Tuesday, so if > > you want to chat about specifics! > > I do know, but unfortunately I'm going to a conference. This is very > pity for me. > > _______________________________________________ > dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii -- Dipl.-Ing. Uwe Köcher Helmut Schmidt Universität Universität der Bundeswehr Fakultät für Maschinenbau, Professur für Numerische Verfahren in den Rechnergestützten Ingenieurwissenschaften. Holstenhofweg 85 22043 Hamburg Tel: +49 40 6541 - 3540 (vorübergehend) Raum: H1 - R.1103 (vorübergehend) Email: [email protected] Web: http://www.hsu-hh.de/mb-mathe _______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
