Benchmarks
Linear algebra frameworks benchmark
IEEE 39 1-year time series
The experiment is to test the time taken by the time series simulation using different linear algebra solvers.
The power flow tolerance is ser to 1e-4.
The time in seconds taken using each of the solvers is:
KLU |
BLAS/LAPACK |
ILU |
SuperLU |
Pardiso |
|
|---|---|---|---|---|---|
Test 1 Test 2 Test 3 Test 4 Test 5 |
82.0306398868561 80.2231616973877 79.5343339443207 80.0667154788971 80.0720291137695 |
82.1049809455872 80.8419146537781 82.3211221694946 82.6606991291046 80.5129723548889 |
81.7956540584564 81.7140426635742 82.7529213428497 82.1418635845184 81.9473338127136 |
82.8895554542542 81.3713464736938 80.9804055690765 80.178496837616 80.0363531112671 |
93.2362771034241 95.2913007736206 92.6268711090088 97.6009163856506 93.3938195705414 |
Average |
80.3853760242462 |
81.6883378505707 |
82.0703630924225 |
81.0912314891815 |
94.4298369884491 |
2869 Pegase 1-week time series
The experiment is to test the time taken by the time series simulation using different linear algebra solvers.
The power flow tolerance is ser to 1e-4.
The time in seconds taken using each of the solvers is:
KLU |
BLAS/LAPACK |
ILU |
SuperLU |
Pardiso |
|
|---|---|---|---|---|---|
Test 1 Test 2 Test 3 Test 4 Test 5 |
2.46547317504882 2.35307431221008 2.40140151977539 2.33513951301574 2.31796050071716 |
2.50752806663513 2.31241440773010 2.42792844772338 2.31270241737365 2.32209181785583 |
2.52735018730163 2.36424255371093 2.46322917938232 2.34093046188354 2.45891189575195 |
2.48413443565368 2.32830643653869 2.46966910362243 2.33488821983337 2.33409214019775 |
2.54547953605651 2.59090781211853 2.46686577796936 2.42691206932067 2.51999592781066 |
Average |
2.37460980415344 |
2.37653303146362 |
2.43093285560608 |
2.39021806716919 |
2.51003222465515 |
So from the light of these tests the solvers are roughly equivalent except the Pardiso one with is worse than the others for these type of simulations.