Speedrun through Splicing Sockets with Sockmap
summary
Shows how to bypass the kernel network path by wiring sockets together with splice() syscall alternatives and sockmap + BPF programs, loading sk_msg and sk_skb probes, attaching via cgroups and using network namespaces and veth pairs. Demo uses sockperf with a pingpong workload: baseline ~5.9 ms falls to ~4.7 ms; a simple Go proxy drops from ~24 μs to ~20 μs when offloaded. Code and scripts live on GitHub; caveats include an intermediate ingress queue that can build up and limited support for some socket types.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 4:02What can Sockmap do
- 9:02What is Sockmap
- 10:50History
- 14:50Practical stuff
- 18:54Getting references
- 23:15Different setups
- 23:25Send to Local
- 30:50Ingress to Erress
- 33:22Proxy
- 40:01Ingress
- 45:14Outro
related talks
24mSpeedrun Through Splicing Sockets with Sockmap
30mOn Sockets and System Calls: Minimizing Context Switches for the Socket API
01h 21mTutorial: Developing High-performance Networking Applications
36mThe black art of network performances
38mVSOCK: From Convenience to Performant VirtIO Communication
21mCo-opting Linux Processes for High-Performance Network Simulation
26mContainer Networking: The Play of BPF and Network NS with different...
31mA Practical, Research‑Backed Introduction To Cloud‑Native Kernel‑Bypass Networking