Initialization scales badly

[Lennart Ochel]

It seems that the initialization scales badly for large systems.

[Lennart Ochel]

This are some time measurements using ScalableTestSuite.Electrical.TransmissionLine.ScaledExperiments.TransmissionLineModelica_N_XXX as scalable model:

	time [s] for
N	pre optimization	initialization	simulation
10	0,037472811	0.041285136	0.01164375
20	0.071433869	0.079462691	0.020607123
40	0.217333943	0.166757507	0.04152687
80	0.432273499	0.454853872	0.164011598
160	1.17030325	2.267705213	0.37920847
320	5.0227649	20.62593407	1.766217402
640	32.85928258	285.9040669	12.47769637

[Francesco Casella]

Some models (including the one mentioned above) in the library do not have fully specified initial conditions. Maybe the problem is that the algorithm that selects them automatically doesn't scale well with the system size. Wheteher this is an intrinsic property of tha problem, or just due to an inefficient implementation of the current algorithm, might be worth investigating.

However, it is likely that for reasons of efficiency, large-scale systems are provided with fully specified initial conditions.

I will do so and commit a new revision of the library asap for testing.

[Francesco Casella]

The new version of the ScalableTestSuite library with fully specified initial conditions is 1.2.0. It was released today on GitHub, should end up automatically in the testsuite tomorrow

[Lennart Ochel]

Replying to lochel:

This are some time measurements using ScalableTestSuite.Electrical.TransmissionLine.ScaledExperiments.TransmissionLineModelica_N_XXX as scalable model:

	time [s] for
N	pre optimization	initialization	simulation
10	0,037472811	0.041285136	0.01164375
20	0.071433869	0.079462691	0.020607123
40	0.217333943	0.166757507	0.04152687
80	0.432273499	0.454853872	0.164011598
160	1.17030325	2.267705213	0.37920847
320	5.0227649	20.62593407	1.766217402
640	32.85928258	285.9040669	12.47769637

I speeded up the initialization for all models that are not over-determined, which is most common.
This are the new time measurements for exactly the same model as above:

	time [s] for
N	pre optimization	initialization	simulation
10	0.037948374	0.018084288	0.011952443
20	0.073691972	0.036116253	0.02039135
40	0.192043867	0.071482219	0.040158798
80	0.428354136	0.192928505	0.128823696
160	1.085886028	0.706097062	0.366486381
320	5.003769767	3.933397942	1.841361654
640	33.40780092	29.00988272	12.99646612

[Lennart Ochel]

The general issue is fixed. We could open a new ticket for over-determined models, since they will still scale badly.

[Martin Sjölund]

Milestone renamed

[Martin Sjölund]

Milestone renamed