The evolution of service mesh with eBPF
summary
Claims that ebpf can bypass iptables and splice sockets to shave latency and raise throughput, validated with Istio perf benchmarks at ~1,000 RPS showing measurable p50/p99 improvements. Uses: ebpf for redirection, low-level telemetry and proxy placement; not used for full L7 logic—retries, circuit breaking and JWT auth remain in Envoy/Istio. Trade-offs: sidecar per pod keeps fine-grained identity, shared-node proxies cut cost but lose isolation. Solo.IO ships this as a flag in Gloo Mesh and documents the approach in a December blog.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:46What is a service mesh
- 6:18Service mesh evolution
- 9:03Sidecar proxy challenges
- 14:17eBPF in service mesh
- 19:12eBPF basics
- 25:12Accelerating proxy traffic
- 30:07Observability with eBPF
- 32:57Optimizing proxy placement
- 36:06Solo.io’s vision
- 38:11Q&A
related talks
41mPanel Discussion - The Future of Service Mesh: Is eBPF a Silver Lining or a Silver Bullet
31mThe Future of Service Meshes: Sidecarless or eBPF?
11mCilium Service Mesh
11mLightning Talk: Clearing the confusion about eBPF and service mesh
25meBPF Strengthens SR-IOV to Be Powerful
19mKeynote: Hello eBPF! Goodbye Sidecars?
34mHow Many Proxies Do You Need?
28mSafe sharing of the network with eBPF