Lightning Talk: Dissecting Service Mesh Overheads
summary
MeshInsight analyzes Istio sidecar overheads by decomposing per-request work into independent components (syscalls, parsing, proxy processing, IPC), builds offline performance profiles and an online predictor, and composes extended car graph models from distributed traces to predict end-to-end latency and CPU overhead. Applied to Alibaba microservices it finds overheads vary by orders of magnitude: TCP vs gRPC with filters shows ~10x difference at p50 and p50 vs p99 spikes up to ~50x. TLS and control-plane costs are not modeled yet.
AI-written summary · about
related talks
08mLightning Talk: A Practical Guide on How to Monitor and Compare Service Mesh...
10mLightning Talk: Optimise Your Mesh With Istio Sidecar
29mYou Probably DON’T Need a Service Mesh!
28mService Meshes: At What Cost?
37mPanel Discussion: Ask Us Anything: Microservices and Service Mesh
27mService Mess to Service Mesh
36mPanel: Ask Me Anything: Microservices and Service Mesh [B]
10mLightning Talk: MeshMark: Service Mesh Value Measurement