[HPGMG Forum] Solver options used for FV results?

Jed Brown jed at jedbrown.org
Tue Aug 19 02:47:18 UTC 2014


Constantinos Evangelinos <cevange at us.ibm.com> writes:

> Do we have a similar list of options etc. for the FE version?
> I got it working but do not know what I should be looking at of the many
> ways I can run it.

You can run

  hpgmg-fe sample -local 100,5e5 -op_type poisson2

This uses Q2 elements and will sample across a range of local sizes
(approximately geometrically spaced between 100 elements/process to 5e5
elements/process).  Choose the largest size to fit in memory on the
machine.  If I know how to report resident memory usage (e.g., getrusage
or Kernel_GetMemorySize), then it will be output after the largest
problem size has been solved.  Per-process memory usage is
asymptotically linear in local problem size.

  Finite Element FAS Performance Sampler on process grid [1 2 2] = 4
  Max memory per MPI rank: 0.010822 GB
  Small Test G[    6    8    8]
  Q2 G[    6    8    8] P[  1  2  2]  7.310e-03 s    5.163538 GF    0.512989 MEq/s
  Large Test G[   96  128  128]
  Q2  0 e_max 1.22e-01(0.0) e_L2 8.73e-02(0.0) r_2 1.20e-08(0.0) G[    3    4    4] L[   3   4   4] P[  1  1  1]
  Q2  1 e_max 2.21e-02(2.5) e_L2 1.21e-02(2.9) r_2 4.73e-03(-18.6) G[    6    8    8] L[   6   4   4] P[  1  2  2]
  Q2  2 e_max 2.72e-03(3.0) e_L2 1.54e-03(3.0) r_2 2.09e-03(1.2) G[   12   16   16] L[  12   8   8] P[  1  2  2]
  Q2  3 e_max 3.56e-04(2.9) e_L2 1.93e-04(3.0) r_2 9.08e-04(1.2) G[   24   32   32] L[  24  16  16] P[  1  2  2]
  Q2  4 e_max 4.49e-05(3.0) e_L2 2.42e-05(3.0) r_2 4.18e-04(1.1) G[   48   64   64] L[  48  32  32] P[  1  2  2]
  Q2  5 e_max 5.57e-06(3.0) e_L2 3.02e-06(3.0) r_2 2.04e-04(1.0) G[   96  128  128] L[  96  64  64] P[  1  2  2]
  
e_L2 should be converging at third order (3.0)
  
  Q2 G[   96  128  128] P[  1  2  2]  1.404e+01 s    9.253325 GF    0.908157 MEq/s

This solve confirms that the largest problem size fits in memory.

  Max memory per MPI rank: 0.570180 GB

This is an approximate high-water mark for the run.

  Starting performance sampling
  Q2 G[    6    8    8] P[  1  2  2]  7.112e-03 s    5.303737 GF    0.526918 MEq/s
  Q2 G[    6    8    8] P[  1  2  2]  7.947e-03 s    4.750241 GF    0.472023 MEq/s
  Q2 G[    6    8    8] P[  1  2  2]  7.177e-03 s    5.254715 GF    0.522151 MEq/s
  Q2 G[    6    8    8] P[  1  2  2]  7.102e-03 s    5.310852 GF    0.527729 MEq/s
  Q2 G[    6    8    8] P[  1  2  2]  8.552e-03 s    4.410079 GF    0.438221 MEq/s
  Q2 G[   12   12   16] P[  1  2  2]  1.716e-02 s    9.270026 GF    1.197999 MEq/s
  Q2 G[   12   12   16] P[  1  2  2]  1.704e-02 s    9.337978 GF    1.207093 MEq/s
  Q2 G[   12   12   16] P[  1  2  2]  1.837e-02 s    8.673426 GF    1.121188 MEq/s
  Q2 G[   12   12   16] P[  1  2  2]  1.841e-02 s    8.660742 GF    1.119549 MEq/s
  Q2 G[   12   12   16] P[  1  2  2]  1.838e-02 s    8.666799 GF    1.120332 MEq/s
  Q2 G[   16   16   24] P[  1  2  2]  3.555e-02 s   10.834136 GF    1.473244 MEq/s

[...]
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 818 bytes
Desc: not available
URL: <https://hpgmg.org/lists/archives/hpgmg-forum/attachments/20140818/07383294/attachment.bin>


More information about the HPGMG-Forum mailing list