Reconstruction benchmark¶
This record compares direct OpenFOAM reconstruction with StackNordic orchestration. It measures reconstruction overhead, numerical parity and cross-distribution compatibility; it does not measure solver performance.
Reporting¶
Repeated measurements report the median and maximum absolute deviation from the median. End-to-end wall time includes process start-up and shutdown. Reconstruction wall time covers the operation reported by StackNordic after package import.
Validation matrix¶
| Case | Source | Runtime | Processor directories | Selection | Runs |
|---|---|---|---|---|---|
| Taylor–Green vortex LES | OpenFOAM.com v2512 | OpenFOAM.com v2512 | 32 | U at 0.011574954 |
1 warm-up + 7 measured |
| Premixed hydrogen LES | OpenFOAM Foundation 12/14 | OpenFOAM.com v2512 | 384 | U at 0.00114 |
1 |
| Premixed hydrogen LES | OpenFOAM Foundation 12/14 | OpenFOAM Foundation 14 | 384 | U at 0.00114 |
1 |
Direct reconstruction¶
The Taylor–Green vortex comparison alternated direct reconstructPar and case.reconstruct() after one warm-up per path.
| Path | End-to-end wall | Reconstruction wall | Peak process-tree RSS |
|---|---|---|---|
Direct reconstructPar |
3.344 ± 0.067 s | 3.344 ± 0.067 s | 1901.0 ± 40.6 MiB |
case.reconstruct() |
5.932 ± 0.091 s | 4.235 ± 0.037 s | 1929.6 ± 52.2 MiB |
The StackNordic reconstruction ratio was 1.267; the end-to-end ratio was 1.774 when Python start-up, package import, case inspection, runtime preflight, compatibility staging and cleanup were included. All fourteen measured outputs had the same SHA-256 digest and were therefore byte-identical.
Command-line validation¶
The installed stacknordic reconstruct command was validated on Roihu with OpenFOAM.com v2512. The 32-directory Taylor–Green vortex case reconstructed U at 0.011574954. Direct and StackNordic runs were alternated after one warm-up per path.
| Path | End-to-end wall | Runs | Output parity |
|---|---|---|---|
Direct reconstructPar |
2.178 ± 0.020 s | 7 | SHA-256 identical |
stacknordic reconstruct |
4.062 ± 0.035 s | 7 | SHA-256 identical |
The CLI ratio was 1.864. Its end-to-end time includes Python start-up, package import, runtime preflight, isolated case staging and cleanup. Roihu job 1569534 performed the measurement on one interactive CPU node; the raw CSV and summary are retained with the project validation artefacts.
Cross-distribution reconstruction¶
| Runtime | Reconstruction wall | Slurm wall | Peak RSS | Compatibility action |
|---|---|---|---|---|
| OpenFOAM.com v2512 | 18.42 s | 23 s | 243 MiB | Generated ESI boundaryProcAddressing in the isolated case view |
| OpenFOAM Foundation 14 | 16.01 s | 19 s | 278 MiB | Used the native Foundation decomposition |
Both 384-rank runs completed successfully, preserved the source case and removed their isolated case views.