Lightning Talk: Clearing the confusion about eBPF and service mesh
summary
eBPF is not Turing-complete at runtime so it excels where work is stateless and early in the stack: L4 packet filtering (IP and Kubernetes pod-label network policy), DDoS mitigation, and kernel-level tracing that yields kernel-level observability for all workloads. Data-path acceleration using XDP and sockmap demonstrably reduces latency and raises throughput for sidecar offload. Complex L7 work—JWT verification, external auth, HTTP/2/3 stream state, retries, circuit breakers—requires user‑space proxies and stateful logic, so Envoy‑style sidecars remain necessary.
AI-written summary · about
chapters (5)
- 0:00Start
- 0:03Service Mesh and Ebpf
- 1:00Can We Use Ebpf and Service Mesh
- 1:25The Ebf Runtime Model
- 7:44Resiliency
related talks
05mLightning Talk: Back to the Future with eBPF
39mThe evolution of service mesh with eBPF
31mThe Future of Service Meshes: Sidecarless or eBPF?
17mLightning Talk: eBPF based UPF (User Plane Function) for 5G
34mHow Many Proxies Do You Need?
32meBPF: BPF kernel infrastructure
32mDemystifying eBPF - eBPF Firewall from scratch
37mEfficient and Flexible Virtual Machine Networking Through eBPF