Code

The current changeset that is being used is 1290:c8607dac4238.

It is located at /home/ehansen/PulsedJet_2.5D/astrobear3.0

Please note that there are changes to problem.f90 in this directory that have not been committed to the repository.

Fiducial Parameters

global.data​

physics.data​

solver.data​

problem.data​

Key Parameters

Resolution: 42X420 + 7 AMR

cfl_vars = .8, .4, .1

qTolerance            = .10,.25,.40,1d30,1d30,1d30,1d30,1d30

iThreaded             = -1

InterpOpts            = 0,0,0,0,0,0,0,0,0,0,0,0,0

buffers                 = 4, 4, 4, 4, 4, 4, 4, 4

MinDensity       = 1d-4[

MinTemp          = 1d-10

iDensityProtect  = 2

iMomentumProtect = 2

iPressureProtect = 2

Parameter Tests

	Fiducial	5 AMR	3 AMR	Buffer increase	Density Refinement
% root steps restarted	0.14%
Time to frame 1	1min
Last frame	50
Time for last frame	4.5 hours
Est. time to completion	12 hours
Number of processors	256
Location	/bamboodata/bliu/Eddie/PulsedJet_2.5D/Fiducial/
Other deviations from Fiducial
Last Frame / movie link

Resolution Tests

The idea here is a bit different than adjusting data file parameters to see what works best. Here, I am simply lowering the number of levels while keeping the number of processors constant. This is to see how having a larger root grid might help performance when running on many processors.

There is density gradient refinement to level 3, and cooling time threshold refinement to level 7 (based on fiducial run). qTolerance changes for each run to keep this consistent. For example, the level 6 run has

qTolerance = .25, .40, 1d30, 1d30, ...

so that there is the same refinement criteria at the same effective resolution. For level 5…

qTolerance = .40, 1d30, 1d30, 1d30, ...

The level 4 and level 3 runs have the same qTolerance…

qTolerance = 1d30, 1d30, 1d30, 1d30, ...

The level 8 run is a little different. Now we want density refinement to level 4, but we have no way of getting the exact same amount of refinement as the level 7 run. I just used this…

qTolerance = .10, .25, .40, .55, 1d30, 1d30, ...

	Fiducial	6 AMR	5 AMR	4 AMR	3 AMR	8 AMR
% root steps restarted	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%
Time to frame 1	2 min	2 min	4 min	10 min		2 min
Last frame	12	12	9	4		13
Time for last frame	1 hr	1 hr	1 hr	1 hr	1 hr	1 hr
Est. time to completion	8.7 hr	9.1 hr	11.3 hr	21.5 hr		7.5 hr
Number of processors	1008	1008	1008	1008	1008	1008
Location	kraken	kraken	kraken	kraken	kraken	kraken
Other deviations from Fiducial	N/A	qTolerance (see above)	qTolerance	qTolerance	qTolerance	qTolerance
Last Frame / movie link

With these runs, I was mainly interested in how many frames I could get in 1 hr, and how the number of levels would affect this and the number of restarts.

As you can see from the table, the restarts seem to have disappeared. Restarts could still be triggered at later frames, but at least this particular set of parameters helped with the first hour of run time. In conclusion, it looks like it will be okay to run with 7 levels (maybe even 8) with these parameters.