Repository navigation
Performance Benchmarking #52
Description
Create benchmarks for solvers, mesh generators, and visualization routines. Integrate these benchmarks into the existing node.js examples. For visualization, implement an export-to-image feature to support the benchmarks.
Tasks
- Collect and store performance metrics, including runtime, peak memory usage, and result accuracy. Regarding results accuracy, we can store a reference solution, for every example, and compare the solution of some random nodes with the reference solution.
- Automate benchmark runs in CI. Extend the current workflow, which currently performs only basic runs.
Activity
nikoscham commentedon Nov 4, 2026
The performance benchmarks are now in a separate repository, FEAScript-benchmark, instead of in the Node.js examples. Each benchmark runs FEAScript and Elmer on identical meshes, on one pinned core, and records for every machine and FEAScript version:
There are two benchmarks so far:
frontal,lusolveandjacobi, linear and quadratic meshes up to 51,681 nodesResults are kept for FEAScript 0.2.0, 0.3.1 and 0.4.0 (RC), with plots of the speed-up by version. They have already guided #121, #120 and #132; for example, the 64×64 cavity went from 145 s to 10 s. The Solvers page of the documentation summarizes the results.
On the two tasks:
tests/regressiondoes.Benchmarks of mesh generation and of plotting, with an export to image, are not covered yet. If needed, they can be opened as new issues in FEAScript-benchmark. Closing this one.