RUNTIME & MEASUREMENT
Cohuno runtime architecture and benchmarks
Persistent workers reuse process initialization across XML operations. Prewarming moves that preparation before business requests; it does not remove the preparation cost.
Internal 17-activity order-fulfilment benchmark
The recorded warm engine median was 6.33 milliseconds across 200 sequential executions. The 95th percentile was 7.61 ms and the 99th percentile 8.26 ms. No additional XML workers were started during those warm executions.
Scope
This test used the Python compatibility runtime in a shared Linux x86-64 sandbox, with simulated external connectors. It excluded API transport, database persistence, and live third-party services. Prewarming two workers took approximately 2.61 seconds and was measured separately. These measurements are not Rust runtime results, a production SLA, a concurrency benchmark, or a comparison with TIBCO or MuleSoft.
The native Rust runtime included in the repository is an experimental candidate and is not established by this benchmark.
Reproduce and inspect
In the repository, run python migration-v3/scripts/benchmark_compat.py --samples 200 --output benchmark.json with the documented Python dependencies installed. Results depend on hardware, engine configuration, payload, and workload.
Inspect the original benchmark JSON
Measure total cost on your workload
Account for infrastructure, persistent worker memory, connector dependencies, operation volume, observability, and support. No percentage savings against another product is established by this test.