Good morning,
Is it possible to use the paralleling computation with the direct method: umfpack? Because Elmer gave me an error when I used this method.
Regards,
Andrea
Paralleling computation with umfpack
-
- Posts: 176
- Joined: 22 Mar 2021, 18:39
- Antispam: Yes
- Location: POLYTHECNIC UNIVERSITY OF TURIN
Paralleling computation with umfpack
POLYTECHNIC UNIVERSITY OF TURIN-DIMEAS
-
- Site Admin
- Posts: 4838
- Joined: 22 Aug 2009, 11:57
- Antispam: Yes
- Location: Espoo, Finland
- Contact:
Re: Paralleling computation with umfpack
Hi Andrea,
No. Umfpack is not MPI parallel library. Of the actively supported direct linear solvers MUMPS is parallel, also on some platforms you could compile with MKL Pardiso.
This is not a limitation by Elmer. It is just that many algorithms are just so much easier to code in serial that the MPI version is often lacking. Direct linear methods belong to this group. Instead, iterative linear solvers are pretty easy to parallelize and as a result there are many implementations available.
-Peter
No. Umfpack is not MPI parallel library. Of the actively supported direct linear solvers MUMPS is parallel, also on some platforms you could compile with MKL Pardiso.
This is not a limitation by Elmer. It is just that many algorithms are just so much easier to code in serial that the MPI version is often lacking. Direct linear methods belong to this group. Instead, iterative linear solvers are pretty easy to parallelize and as a result there are many implementations available.
-Peter
-
- Posts: 176
- Joined: 22 Mar 2021, 18:39
- Antispam: Yes
- Location: POLYTHECNIC UNIVERSITY OF TURIN
Re: Paralleling computation with umfpack
Hi Peter,raback wrote: ↑20 Jul 2021, 00:35 Hi Andrea,
No. Umfpack is not MPI parallel library. Of the actively supported direct linear solvers MUMPS is parallel, also on some platforms you could compile with MKL Pardiso.
This is not a limitation by Elmer. It is just that many algorithms are just so much easier to code in serial that the MPI version is often lacking. Direct linear methods belong to this group. Instead, iterative linear solvers are pretty easy to parallelize and as a result there are many implementations available.
-Peter
Thanks for your answer. Which platform do they support MKL Pardiso? I tried to use a paralleling computation just to speed up the simulations I am running. Unfortunately, I have been noting that the latter ones have divergence problems. Indeed, I posted a thread in the section "Elmer solver"
Regards,
Andrea
POLYTECHNIC UNIVERSITY OF TURIN-DIMEAS