Session

Precision timing in Linux networking has largely been limited to control-plane timestamping, where packets are observed but not controlled, and typically at rates below modern line speeds. Bringing timestamping into the data plane at line rate is itself non-trivial: queueing, batching, and driver behavior introduce variability that undermines timestamp fidelity even with hardware assist. We characterize these effects at high throughput and present techniques for enabling accurate, high-rate data-plane timestamping, closing a long-standing gap in the stack.

However, visibility alone is insufficient. Linux can measure time but does not yet reliably enforce it in the transmit path. This talk explores how SO_TXTIME shifts timing from observation to control, enabling applications to explicitly schedule packet transmission. While powerful, software-based scheduling alone struggles to scale to modern link rates and tight jitter bounds. To complete the model, we propose the use of hardware-timed pause semantics, enabling precise, low-jitter control at line rate. Together, these approaches move toward deterministic, time-aware data planes.