This example, together with BuildingSystems.Fluid.Examples.Performance.Example7, illustrates the overhead generated by divisions by parameters. See Jorissen et al. (2015) for a complementary discussion.
Running the following commands allows comparing the CPU times of the two models, disregarding as much as possible the influence of the integrator:
simulateModel("BuildingSystems.Fluid.Examples.PerformanceExamples.Example6", stopTime=100, numberOfIntervals=1, method="Rkfix4", fixedstepsize=0.001, resultFile="Example6"); simulateModel("BuildingSystems.Fluid.Examples.PerformanceExamples.Example7", stopTime=100, numberOfIntervals=1, method="Rkfix4", fixedstepsize=0.001, resultFile="Example7");
Comparing the CPU times indicates a speed improvement of 56%.
This difference almost disappears when adding annotation(Evaluate=true)
to R
and C
.
In dsmodel.c
we find:
DynamicsSection W_[2] = divmacro(X_[0]-X_[1],"T[1]-T[2]",DP_[0],"R"); F_[0] = divmacro(W_[1]-W_[2],"Q_flow[1]-Q_flow[2]",DP_[1],"C");
This suggests that the parameter division needs to be handled during each function evaluation, probably causing the increased overhead.
The following command allows comparing the CPU times objectively.
simulateModel("BuildingSystems.Fluid.Examples.Performance.Example6", stopTime=100, numberOfIntervals=1, method="Euler", fixedstepsize=0.001, resultFile="Example6");
See Jorissen et al. (2015) for a discussion.