We want to be able to benchmark our server implementations to compare implementation techniques and find performance regressions.
For now, there is a simple benchmark setup in run.sh, which should be run via make benchmark from the project root.
It depends solely on dnspyre.
For each implementation (basic, threaded-4, tokio), the benchmark collects the following artefacts:
- DNS query response times histograms in
raw.csv - the
stdoutstream from the benchmarking tooldnspyre - latency and throughput plots in
graphs-<timestamp>
These results are not qualitative benchmarks and are not run in hygenic benchmarking environments, but they should give a basic stepping stone
Each commit on main triggers a benchmark step after successful compilation & linting.
The benchmark artefacts are zipped and attached to the respective CI jobs for inspection.
The goal is to automatically embed these artefacts within a webpage to automatically detect performance regressions.