Network-model performance comparison
The topology and model-name refactor was compared with the frozen source at commit 28553eb6e2cb36df5f86c6d5194dbd4232ea1812. The current measurement used the uncommitted refactor working tree based on the same commit.
Host and method
- Date: 2026-08-19
- Julia: 1.12.6
- Operating system: Linux 5.14.0-687.33.1.el98.x8664
- Processor: Intel Core i9-13950HX
- Fixture: three parallel scalar impedance elements between one retained node and ground
- Frequency vector: 400 logarithmically spaced points from 1 Hz to 10 kHz
- Calculation: passive
make_y_edgeevaluation from the linearized network model - Warm-up: two complete evaluations before measurement
- Samples: 200 execution-time and allocation measurements
- Reported statistic: median
The baseline used the former pin grammar and LinearizedAdmittanceNetwork. The refactored case used equivalent named topology rows and NetworkModel. Component values, node selection, frequency axis, Julia process settings, and host were otherwise identical.
Results
| Source | Median time | Median allocated bytes |
|---|---|---|
| Frozen baseline | 0.9135885 ms | 2,305,856 B |
| Refactored working tree | 0.9194500 ms | 2,305,776 B |
| Relative change | +0.64% | −0.0035% |
The measured time increase is below the 10% acceptance limit. Allocated bytes decreased by 80 bytes and therefore satisfy the 5% allocation limit.
These measurements describe warmed admittance evaluation on this host. CI job duration includes dependency loading, compilation, solver fixtures, and runner scheduling and is not a substitute for this comparison.