Benchmarking Journey Through Service Meshes
Talkat Linkerd Day Europe 2025 · on Tuesday, 2025-04-01· at 17:15· Level 3 | ICC Capital Suite 14-16·
summary
Benchmarks run on a GKE cluster with wrk2 driving the emoji app (3 services, gRPC) at 20/200/2000 RPS, reinstalling deployments between runs and enabling mTLS (ambient gets a waypoint proxy for layer‑7 fairness). Results report median plus P90/P95/P99 latencies: at 20 RPS everything matches baseline, at 200 RPS Linkerd and ambient track baseline within milliseconds while sidecar meshes show higher P99 spikes, and at 2000 RPS Linkerd stays slightly faster while ambient (waypoint enabled) closes much of the 2021 gap.
AI-written summary · about
related talks
27mA Service Mesh Benchmark You Can Trust
08mLightning Talk: A Practical Guide on How to Monitor and Compare Service Mesh...
53mDo you need a Service Mesh?
56mObservability in Action for your Service Mesh
29mDebugging an app with a service mesh
25mUsing Istio to Build Better Service Level Objectives
37mAutoscaling Multi-Cluster Observability with Thanos and Linkerd
29mMeasuring K8s Network Performance