A Service Mesh Benchmark You Can Trust
summary
Runs a live benchmark on a kind cluster using bare‑metal machines and a five‑year AMD CPU to show real costs instead of synthetic claims. Uses K6 client, Fortio backend, Prometheus and Grafana dashboards and default Helm Istio install in ambient mode. Measures ~170 µs baseline, ~750 µs with sidecar L7, ~150–350 µs ambient L4, and ~700 µs ambient with waypoint L7; scripts and reproduction steps are published and the demo completes successfully.
AI-written summary · about
chapters (13)
- 0:00Service mesh benchmarks intro
- 2:04Benchmarking service meshes is hard
- 3:31Test setup and tools
- 6:22Deploy benchmark cluster
- 9:36Run baseline test
- 13:05Sidecar performance
- 15:13Layer 7 policy tests
- 17:41Ambient layer 4 tests
- 20:07Ambient layer 7 with waypoints
- 22:15Why migrate to ambient
- 24:02Benchmark results
- 25:28Replicate the benchmark
- 26:51Q&A
related talks
06mBenchmarking Journey Through Service Meshes
08mLightning Talk: A Practical Guide on How to Monitor and Compare Service Mesh...
34mIstio's Ambient Mesh: The Real Cost of Sidecar-less Tracing
25mUsing Istio to Build Better Service Level Objectives
09mLightning Talk: Dissecting Service Mesh Overheads
56mObservability in Action for your Service Mesh
26mAre You Sure About Your Mesh Performance?
53mDo you need a Service Mesh?