Tailscale describes packet-handling optimizations achieving 5% speed-up

2 Yesterday · 1d ago · 1 post · 1 source · development 2 of 3

Tailscale detailed how on Linux and Android, it now avoids copying small packets into separate 64 KiB buffers, instead identifying packet boundaries within a single large read. The approach keeps small packets small in memory and reduces copying overhead, yielding a roughly 5% speed-up in many network configurations.