Some numbers to chew. The case is the duct above where various parameters are either provided by CoolProp (ML) or constant (C) or - in case of density - by own formula (UL) - Bon appetit, and thank you for your being so helpful.
ML = Material Law
C = Constant
UL = User Law (density = rho0*(293.15/temperature);)
Air, k-e, 50 iterations, single CPU, two threads.
Matrix bandwidth for volume mesh : 61491
Matrix profile/lines for volume mesh : 22235
Dens Visc SpcH TCnd
./RESU/20231030-1941 ML ML ML ML
Calculation time summary:
User CPU time: 331.540 s
System CPU time: 1.464 s
Elapsed time: 204.167 s
CPU / elapsed time 1.631
Total elapsed time for CoolProp (TTSE&HEOS): 46.584 s
./RESU/20231030-1946 ML ML ML C
Calculation time summary:
User CPU time: 320.180 s
System CPU time: 1.240 s
Elapsed time: 190.581 s
CPU / elapsed time 1.687
Total elapsed time for CoolProp (TTSE&HEOS): 37.175 s
./RESU/20231030-1950 ML ML C C
Calculation time summary:
User CPU time: 308.492 s
System CPU time: 1.340 s
Elapsed time: 180.402 s
CPU / elapsed time 1.717
Total elapsed time for CoolProp (TTSE&HEOS): 26.870 s
./RESU/20231030-1953 ML C C C
Calculation time summary:
User CPU time: 299.348 s
System CPU time: 1.084 s
Elapsed time: 168.838 s
CPU / elapsed time 1.779
Total elapsed time for CoolProp (TTSE&HEOS): 17.286 s
./RESU/20231030-1958 C C C C
Calculation time summary:
User CPU time: 237.760 s
System CPU time: 3.864 s
Elapsed time: 130.127 s
CPU / elapsed time 1.857
Total elapsed time for CoolProp (HEOS): 0.000 s
./RESU/20231030-2019 UL C C C
Calculation time summary:
User CPU time: 299.888 s
System CPU time: 7.384 s
Elapsed time: 180.888 s
CPU / elapsed time 1.699
Total elapsed time for CoolProp (HEOS): 0.000 s
Edit: Added info for "Total elapsed time for CoolProp (TTSE&HEOS)"
Last edited by tpa on Tue Oct 31, 2023 6:04 pm, edited 1 time in total.
Thanks for the data. Since your results are quite different between a constant and a user law, I assume varying density also has some impact on the convergence behavior, making the comparison a bit more tricky, but it is good to know that although using CoolProp adds some computational cost, it is now "reasonabe" compared to the ×10 we had in older versions (recomputing everything at all cells instead of using CoolProp's tabulation mechanism).
1. More variables (opposed to constants) to be calculated require more time
2. The difference between the Material law ML and User Law (UL) for density is in significant.
3. Biggest difference seems to be between all constant or not all constant.
Maybe the call to external routines in general is expensive (program logistics)
You can actually get the time required specifically for calls to CoolProp in the "Physical property computations" sectoon of the "performance.log" file.
Run 1958 1953 1950 1946 1941 2019
Elapsed 130.13 168.84 180.40 190.58 204.17 180.89
CoolPr. 0.00 17.29 26.87 37.18 46.58 0.00
Prop% 0.0% 10.2% 14.9% 19.5% 22.8% 28.1%
Rho C ML ML ML ML UL
Visc C C ML ML ML C
SpHt C C C ML ML C
ThCd C C C C ML C
Incr. 0.00 38.71 11.56 10.18 13.59 50.76
Incr.% 0.00% 10.24% 4.66% 4.61% 3.31% 28.06%
Incr. is the added amount of time spent in CoolProp compared to the run to the left (which has one calculated property less) except 2019 which compares to 1958
Incr.% is the absolute difference in the Prop% compared to the run to the left (which has one calculated property less) except 2019 which compares to 1958
By comparing percentages is to reduce the effect of differences in iteration among the different runs
It seems cheaper to use CoolProp density function than the user formula. As 1950 and 2019 are both around 180s elapsed time it seems that one can have the Viscosity calculation for free.
Each additional value besides density (viscosity, specific heat, or thermal conductivity) costs 3-5% for this problem