Any reports on the processorAgglomerator feature ?

louisgag · July 16, 2015, 05:31

Hello everyone,

I quickly tested the new feature of processorAgglomerator and thought I'd share what I found.

The feature has been available for a while now as posted there: http://www.openfoam.org/version2.3.0/parallel.php but I haven't seen any mention of it in the forums here.

I've tested very quickly (doing only ~25 timesteps) the masterCoarsest and procFaces options and have not found positive changes in the solution speed.

On a 1.7M cell case with pimpleFoam without processorAgglomerator and 10 procs i get a rougly 3.5 times faster solution than on a single processor.

I do get a change in the
nIntFaces/nCells but not in the solution speed. With the nAgglomeratingCells 10500; I'll get

Code:

                              nCells       nFaces/nCells         nInterfaces    nIntFaces/nCells     profile
   Level  nProcs         avg     max         avg     max         avg     max         avg     max         avg
   -----  ------         ---     ---         ---     ---         ---     ---         ---     ---         ---
       0      10      172927  174538       3.023   3.043           6       9     0.03318  0.0408   7.906e+08
       1      10       85177   86196       3.637   3.781           6       9     0.05571 0.06825   2.499e+08
       2      10       41989   42612       4.396   4.641           6       9     0.08784  0.1086   8.005e+07
       3      10       20395   20848       5.159   5.411           6       9      0.1379  0.1679   2.511e+07
       4       5       20134   20426       5.589   5.795         3.6       4      0.1386  0.1686   1.699e+07
       5       2       24366   29273       5.838   5.886           1       1     0.06463 0.07765   1.756e+07
       6       1       23939   23939       5.927   5.927           0       0           0       0   1.297e+07
       7       1       11572   11572       5.891   5.891           0       0           0       0    4.01e+06
       8       1        5680    5680       5.901   5.901           0       0           0       0   1.184e+06
       9       1        2763    2763       5.852   5.852           0       0           0       0   3.833e+05
      10       1        1349    1349       5.894   5.894           0       0           0       0    1.18e+05
      11       1         658     658       5.912   5.912           0       0           0       0   3.962e+04
      12       1         320     320       5.822   5.822           0       0           0       0   1.269e+04
      13       1         152     152       5.783   5.783           0       0           0       0        4327

and with none I get,

Code:

    processor agglomerator : noneGAMGProcAgglomeration

                              nCells       nFaces/nCells         nInterfaces    nIntFaces/nCells     profile
   Level  nProcs         avg     max         avg     max         avg     max         avg     max         avg
   -----  ------         ---     ---         ---     ---         ---     ---         ---     ---         ---
       0      10      172927  174538       3.023   3.043           6       9     0.03318  0.0408   7.906e+08
       1      10       85177   86196       3.637   3.781           6       9     0.05571 0.06825   2.499e+08
       2      10       41989   42612       4.396   4.641           6       9     0.08784  0.1086   8.005e+07
       3      10       20395   20848       5.159   5.411           6       9      0.1379  0.1679   2.511e+07
       4      10       10067   10306       5.556   5.833           6       9      0.2033  0.2487   7.405e+06
       5      10        4873    5023       5.734   6.009           6       9       0.291  0.3532    2.21e+06
       6      10        2393    2478       5.722   5.972           6       9      0.4096   0.509   5.955e+05
       7      10        1157    1207       5.606   5.868           6       9      0.5675  0.7287   1.822e+05
       8      10         568     592       5.497   5.773           6       9      0.8041   1.087    4.67e+04
       9      10         276     287       5.289   5.543           6       9       1.122   1.543   1.492e+04
      10      10         134     140       5.073   5.391           6       9       1.637   2.243        3757
      11      10          65      68       4.754   5.179           6       9       2.311   3.288        1224
      12      10          32      33       4.178    4.75           6       9       3.286   4.656         304
      13      10          15      16       3.364   3.938           6       9       4.835     7.2        85.4

Anyone has observed better results ?

Regards,

-Louis

tomf · July 17, 2015, 06:19

Hi Louis,

When it came out we have also started to look into it. Like you we have not performed a complete benchmark. At least we have not noticed a slowdown, but I guess it is also case dependent. It seems like there is a small gain in scalability for some problems, but we should do a more complete benchmark for production runs. I would not expect a huge improvement however.

Regards,
Tom

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Automatically export all reports in batch	f-w	STAR-CCM+	9	November 17, 2022 04:14
Fluent result reports	Jiricbeng	OpenFOAM Post-Processing	3	July 7, 2014 03:57
Net value reports in Fluent	arizwan	FLUENT	1	July 30, 2013 03:14
Restrict Access to LANL Reports due to 911?	Zi-Wei Chiou	Main CFD Forum	3	November 7, 2001 11:18
Heat Flux Calculation under REPORTS	Alberto Schroth	FLUENT	0	May 16, 2000 08:19

July 17, 2015, 06:19		#2
tomf Senior Member Tom Fahner Join Date: Mar 2009 Location: Breda, Netherlands Posts: 638 Rep Power: 32	Hi Louis, When it came out we have also started to look into it. Like you we have not performed a complete benchmark. At least we have not noticed a slowdown, but I guess it is also case dependent. It seems like there is a small gain in scalability for some problems, but we should do a more complete benchmark for production runs. I would not expect a huge improvement however. Regards, Tom