Hi Victor, > However, can you advise me what iterative solvers D2 has natively that > scale to large (or even modest) single-node core counts?
I think Wolfgang's answer is very good. If you need a solver that
scales, MPI is usually used (see PETSc or Trilinos) and that case you do
not care much if use one node or more nodes.
> Also: I can not find information on how to measure performance. Do you
> have a petsc-style log view that tells me how much time was spent in
> each library routine & the corresrponding flop rate?
There is an additional petsc command line option that outputs the
internal PARDISO perfomance analysis. It is
-mat_mkl_pardiso_68 1
Here is an example output ( a million degrees of freedom on 20 Threads):
=== PARDISO: solving a real nonsymmetric system ===
0-based array is turned ON
PARDISO double precision computation is turned ON
METIS algorithm at reorder step is turned ON
Scaling is turned ON
Matching is turned ON
Summary: ( reordering phase )
================
Times:
======
Time spent in calculations of symmetric matrix portrait (fulladj): 0.543464 s
Time spent in reordering of the initial matrix (reorder) : 0.578133 s
Time spent in symbolic factorization (symbfct) : 1.705050 s
Time spent in data preparations for factorization (parlist) : 0.010526 s
Time spent in allocation of internal data structures (malloc) : 0.035932 s
Time spent in matching/scaling : 1.293615 s
Time spent in additional calculations : 1.807056 s
Total time spent : 5.973776 s
Statistics:
===========
Parallel Direct Factorization is running on 20 OpenMP
< Linear system Ax = b >
number of equations: 925696
number of non-zeros in A: 73808896
number of non-zeros in A (%): 0.008613
number of right-hand sides: 1
< Factors L and U >
number of columns for each panel: 72
number of independent subgraphs: 0
< Preprocessing with state of the art partitioning metis>
number of supernodes: 46229
size of largest supernode: 6081
number of non-zeros in L: 402294842
number of non-zeros in U: 375995776
number of non-zeros in L+U: 778290618
=== PARDISO is running in In-Core mode, because iparam(60)=0 ===
Percentage of computed non-zeros for LL^T factorization
1 % 2 % 3 % 4 % 5 % 7 % 8 % 9 % 10 % 11 % 12 % 13 % 15 % 16 % 17
% 18 % 19 % 20 % 21 % 22 % 23 % 24 % 25 % 26 % 27 % 28 % 29 % 30 %
31 % 32 % 33 % 34 % 35 % 36 % 37 % 38 % 39 % 40 % 41 % 42 % 43 %
44 % 45 % 46 % 47 % 48 % 49 % 50 % 51 % 52 % 53 % 54 % 55 % 56 %
57 % 58 % 59 % 60 % 61 % 62 % 63 % 64 % 65 % 66 % 67 % 68 % 69 %
70 % 71 % 72 % 73 % 74 % 75 % 76 % 77 % 79 % 81 % 82 % 84 % 85 %
86 % 87 % 88 % 90 % 92 % 93 % 95 % 97 % 98 % 99 % 100 %
Best,
Nils
Victor Eijkhout <[email protected]> writes:
> Nils,
>
> I may try Pardiso.
>
> However, can you advise me what iterative solvers D2 has natively that scale
> to large (or even modest) single-node core counts?
>
> Also: I can not find information on how to measure performance. Do you have a
> petsc-style log view that tells me how much time was spent in each
> library routine & the corresrponding flop rate?
>
> V.
--
Nils Schween
Phone: +49 6221 516 557
Mail: [email protected]
PGP-Key: 4DD3DCC0532EE96DB0C1F8B5368DBFA14CB81849
Max Planck Institute for Nuclear Physics
Astrophysical Plasma Theory (APT)
Saupfercheckweg 1, D-69117 Heidelberg
https://www.mpi-hd.mpg.de/mpi/en/research/scientific-divisions-and-groups/independent-research-groups/apt
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups
"deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion visit
https://groups.google.com/d/msgid/dealii/87h5o3ua44.fsf%40mpi-hd.mpg.de.
smime.p7s
Description: S/MIME cryptographic signature
