Real-Time Tuning

Adora provides optional real-time features for latency-sensitive robotics deployments.

Quick Start

# Start daemon with real-time profile (mlockall + SCHED_FIFO)
sudo adora daemon --rt

# Control worker threads
adora daemon --worker-threads 4

# Pin to specific CPU cores
taskset -c 2,3 adora daemon --rt

What `--rt` Does

mlockall(MCL_CURRENT | MCL_FUTURE) — pins all memory, prevents page faults
SCHED_FIFO priority 50 (Linux only) — real-time scheduling for the main thread
Requires CAP_SYS_NICE + CAP_IPC_LOCK capabilities

Per-Node CPU Affinity

Instead of pinning the entire daemon with taskset, you can pin individual nodes to specific CPU cores via the dataflow YAML:

- id: controller
  path: ./controller
  cpu_affinity: [2, 3]

- id: sensor
  path: ./sensor
  cpu_affinity: [4, 5]

The daemon calls sched_setaffinity on the spawned process. This is Linux only; on other platforms the field is silently ignored.

Combine with --rt for best results: the daemon gets real-time scheduling while each node is pinned to dedicated cores, avoiding contention.

Full Guide

See the comprehensive Real-Time Tuning Guide for:

Linux kernel tuning (CPU governor, PREEMPT_RT, boot parameters)
Process-level tuning (CPU affinity, memory locking, thread priority)
Systemd service configuration
Docker and Kubernetes deployment
Zenoh transport tuning
Benchmarking tips