Distillation

Date: Sep 17, 2026

• 🟡 (08:50) User asked what can or should be done about Tailscale’s warning that a UDP forwarding rule is not optimized for br0, referencing https://tailscale.com/docs/reference/best-practices/performance#ethtool-configuration.
• 🔴 (08:50) User stated that direct connections always result in lower latency and higher throughput.
• 🟡 (08:50) User supplied Tailscale performance guidance last validated Dec 4, 2025: Linux subnet routers and exit nodes using Tailscale 1.54+ with Linux kernel 6.2+ can improve UDP throughput through transport-layer offloads; recommended configuration is NETDEV=$(ip -o route get 8.8.8.8 | cut -f 5 -d " ") followed by sudo ethtool -K $NETDEV rx-udp-gro-forwarding on rx-gro-list off.
• 🟡 (08:50) User supplied Tailscale’s persistence method for distributions using networkd-dispatcher: verify with systemctl is-enabled networkd-dispatcher; create /etc/networkd-dispatcher/routable.d/50-tailscale using printf '#!/bin/sh\n\nethtool -K %s rx-udp-gro-forwarding on rx-gro-list off \n' "$(ip -o route get 8.8.8.8 | cut -f 5 -d " ")" | sudo tee /etc/networkd-dispatcher/routable.d/50-tailscale; run sudo chmod 755 /etc/networkd-dispatcher/routable.d/50-tailscale; test with sudo /etc/networkd-dispatcher/routable.d/50-tailscale and test $? -eq 0 || echo 'An error occurred.'.
• 🟡 (08:50) Assistant stated it would inspect the live warning, interfaces, offload state, and host network configuration; compare them with current Tailscale guidance; and avoid privileged changes until the root cause and durable configuration path were clear.
• 🟡 (08:50) Live host uses Tailscale 1.102.3, commit 9329c3677031109ff6d0b80abee0cddc8f35ff6f, long version 1.102.3-t9329c3677-ga522f65e9, other commit a522f65e92d54b594fe72e842afc9572bd9cf0b8, and Go go1.26.6 (tailscale/go 7275f792d4).
• 🟡 (08:50) Tailscale device labs has address 100.69.65.125, hostname labs.sheep-fir.ts.net, runs Linux, is idle, and offers an exit node. Peer desktop-83gep91 (100.102.146.109, Windows) was active via direct endpoint 192.168.50.78:41641 with tx 43211224 and rx 2664056; byxel10pf (100.78.78.55, Android) was offline for 13h; glkvm (100.83.76.82, Linux) was offline for 115d.
• 🟡 (08:50) Tailscale Funnel was on. Connectivity report showed UDP available; public IPv4 185.253.124.189:39670; public IPv6 [2a07:244:48:bd00:ba85:84ff:fe9c:ea2a]:35733; MappingVariesByDestIP: false; no port mapping; nearest DERP London. Gateway/self-IP monitor reported gw=192.168.50.1 self=192.168.50.13.
• 🟡 (08:50) DERP latencies were: London 11.1ms, Paris 15.2ms, Amsterdam 20.1ms, Frankfurt 23.8ms, Nuremberg 24.3ms, Warsaw 32.6ms, Madrid 34.6ms, Helsinki 50.9ms, New York City 73.5ms, Ashburn 80ms, Toronto 88.2ms, Chicago 94.5ms, Miami 107ms, Dallas 108.4ms, Denver 114.1ms, Dubai 117ms, Los Angeles 132.1ms, Seattle 133.1ms, San Francisco 137.2ms, Nairobi 139.3ms, Bengaluru 143.4ms, Johannesburg 163.1ms, Honolulu 186.6ms, Singapore 188.1ms, Hong Kong 196.3ms, São Paulo 199.3ms, Tokyo 233.6ms; Sydney had no reported latency.
• 🟡 (08:50) ip -o route get 8.8.8.8 resolved through br0: 8.8.8.8 via 192.168.50.1 dev br0 src 192.168.50.13 uid 1000.
• 🟡 (08:50) Host runs Linux kernel 6.8.0-117-generic. Interface br0 was up with MTU 1500, MAC b8:85:84:9c:ea:2a, one TX queue, one RX queue, gso_max_size 64000, gso_max_segs 64, tso_max_size 64000, tso_max_segs 64, and gro_max_size 65536.
• 🟡 (08:50) ethtool was not installed; shell returned /usr/bin/bash: line 1: ethtool: command not found, preventing inspection or configuration of rx-udp-gro-forwarding and rx-gro-list.
• 🟡 (08:50) Assistant diagnosed the Tailscale warning as valid: labs is a Linux 6.8 exit node, its default routed interface is bridge br0, and missing ethtool means the recommended offload flags cannot be inspected or set. Assistant stated persistence should attach to the interface lifecycle rather than rely on a one-time command.
• 🟡 (08:51) Bridge br0 has three active member interfaces: Ethernet enp2s0 with MAC b8:85:84:9c:ea:2a, Wi-Fi wlp3s0 with MAC 2c:98:11:06:75:e3, and Wi-Fi wlx5ce9311c6734 with MAC 5c:e9:31:1c:67:34.
• 🟡 (08:51) br0 has DHCP IPv4 192.168.50.13/24, global IPv6 2a07:244:48:bd00:ba85:84ff:fe9c:ea2a/64, ULA IPv6 fda0:affb:c9e5:1ec3:ba85:84ff:fe9c:ea2a/64, and link-local IPv6 fe80::ba85:84ff:fe9c:ea2a/64. IPv4 default gateway is 192.168.50.1; IPv6 default gateway is fe80::ca7f:54ff:fe91:8600; advertised IPv6 MTU is 1492.
• 🟡 (08:51) br0 is managed by systemd-networkd using /usr/lib/systemd/network/99-default.link and /etc/systemd/network/30-br0.network; its state is routable (configured), online state is online, type/kind is bridge, driver is bridge, activation policy is up, and it is required for online operation.
• 🟡 (08:51) networkd-dispatcher was enabled and active; systemd-networkd was enabled and active. NetworkManager tooling was absent: nmcli: command not found.
• 🟡 (08:51) networkctl reported br0 speed as 100Mbps, DNS servers 1.1.1.1, 1.0.0.1, 192.168.50.1, and 2a07:244:48:bd00::1, with search domain sybrli.