Tailscale details upcoming throughput and memory improvements for app connectors, subnet routers, and exit nodes

Ongoing Performance Investments

Tailscale has consistently invested in improving its data plane performance. Previous efforts included increasing TCP throughput on Linux, achieving over 10Gb/s with wireguard-go on bare metal, and leveraging segmentation offloads for UDP-based applications. These improvements have made Tailscale suitable for performance-sensitive workloads such as continuous integration, remote development, and heavy data transfer.

Upcoming Throughput Enhancements

Tailscale plans to introduce multi-queue technology in the second half of 2026. This will specifically boost throughput for app connectors, subnet routers, and exit nodes, further enhancing network performance across various use cases.

Immediate Client-Side Optimizations

New stable client releases will include immediate improvements to throughput and memory overhead. A key optimization involves how Tailscale handles network packets, particularly on Linux and Android. Previously, even small 1 KiB packets were copied into a 64 KiB buffer, leading to inefficiency. The updated approach allows Tailscale to process packets in place, identifying their start and end points without unnecessary copying, which significantly reduces memory overhead and improves throughput.

Addressing Performance Tooling

Beyond core network performance, Tailscale also aims to solve existing performance tooling issues for its customers. This indicates a broader focus on not just the underlying technology but also the user experience and ability to monitor and troubleshoot performance effectively.