O-DU

The O-DU is the middle piece of the disaggregated base station, and it sits where the timing gets tight. It owns RLC, MAC, and the upper part of the physical layer (high-PHY) — scheduling, HARQ, segmentation, the things that have to react within milliseconds or less. That hard timing is exactly why it can't be pushed all the way back to a central data centre like the O-CU; it has to stay reasonably close to the radio.

Above it, the O-DU connects to the O-CU over F1. Below it, the open fronthaul to the O-RU is the strict one — the 7.2x split leaves low-PHY on the radio, so the O-DU has to deliver frequency-domain samples on time, every slot, which is why fronthaul latency and jitter budgets are measured in microseconds. On the job the O-DU usually runs on a COTS server with hardware acceleration (a lookaside or inline accelerator card) handling the heavy PHY math, since pure software struggles to keep up at scale.