This version is under construction, please use an official release version

Engineering Reference

This section documents how every number in the preceding pages was measured: the definition of a capacity ceiling, the per-test procedures, the distress signals that stop a run, and the cluster tuning the numbers depend on.

  • Ceiling methodology: the definition of a capacity ceiling, the measurement loop, functional verification, and the distress probes that stop a run.
  • Degradation methodology: why tenant bundles are used, the canary topology, the measurement statistics, and the stop thresholds.
  • Per-test method cards: for each capability, what the test creates, what “programmed” means, the parameters used, the stop trigger, and the result.
  • Cluster tuning and limits: the tuning baseline the numbers depend on, the first binding factor per capability, and the interpretation caveats.
  • Glossary: the network control-plane terms used throughout.