We’re making Tailscale faster
Tailscale describes several data-plane changes: Linux/Android packet-buffer reuse that the company says improved speed about 5% in many configurations, writev to reduce copies, and a planned multi-queue pipeline for subnet routers, app connectors, and exit nodes. Netmap caching can also let previously connected devices start sending over the data plane much sooner when the control plane is slow, though it has storage and large-tailnet tradeoffs and is still rolling out.
The Hacker News discussion pushed back on the assumption that moving WireGuard into the kernel automatically makes it faster; Tailscale’s cofounder pointed to userspace optimizations and DPDK as relevant tradeoffs. Commenters also raised a separate user-control complaint about Android VPN exclusions, so the thread was not uniformly about the performance work.