Skip to content

Real-time CPU

Linux real-time (RT) scheduling is critical for applications requiring deterministic latency, such as audio processing, robotics, or industrial control systems. This section covers RT kernel patches, scheduler configuration, and techniques to minimize latency for high-priority tasks.

Understanding Real-Time Scheduling Requirements

Real-time tasks must execute within strict deadlines, often requiring exclusive CPU access and predictable scheduling. Linux provides mechanisms to prioritize such tasks over non-RT workloads, but this requires specific kernel configurations and careful tuning.

Kernel Patches for Real-Time Support

Modern Linux kernels include PREEMPT_RT patches, which replace the standard kernel preemption model with a more aggressive, low-latency approach. Two variants exist:
- CONFIG_PREEMPT_RT: Lightweight patch for minimal latency improvements.
- CONFIG_PREEMPT_FULL: Full patch with deeper changes, suitable for strict real-time requirements.

Enabling these patches requires recompiling the kernel, as they are not part of standard distributions.

Example:

# Check if PREEMPT_RT is enabled in the kernel  
grep PREEMPT_RT /boot/config-$(uname -r)

Real-Time Schedulers and Policies

Linux supports multiple scheduling policies for real-time tasks:

1. SCHED_FIFO (Fixed-Priority)

  • Prioritizes tasks based on user-defined priorities (1–99).
  • Tasks run until they block or yield the CPU.
  • Use case: Critical tasks with strict deadlines.

2. SCHED_RR (Round-Robin)

  • Similar to SCHED_FIFO but with time slices.
  • Less common for hard real-time systems.

3. SCHED_DEADLINE

  • Advanced scheduler for deadline-based tasks.
  • Requires the CONFIG_SCHED_DEADLINE kernel option.
  • Optimizes for both latency and throughput.

Example:

# Set a task's scheduling policy and priority  
sudo chrt -t 99 -p 1234 <pid>  
# Replace <pid> with the process ID

Latency Optimization Techniques

To minimize latency, adjust kernel parameters and system settings:

1. RT Period and Runtime

  • sched_rt_period_us: Defines the time window for RT tasks (e.g., 500000 µs = 500 ms).
  • sched_rt_runtime_us: Maximum CPU time allocated to RT tasks within the period.

Example:

# Set RT period and runtime (requires root)  
echo 500000 > /proc/sys/kernel/sched_rt_period_us  
echo 250000 > /proc/sys/kernel/sched_rt_runtime_us

2. CPU Affinity

Bind RT tasks to specific CPUs to reduce context switching.

Example:

# Assign process 1234 to CPU 0  
sudo taskset -c 0 -p 1234

3. Disable Non-Critical Services

Reduce background processes using:

sudo systemctl disable --now bluetooth avahi-daemon NetworkManager

Monitoring and Verification

Use tools to measure latency and validate configurations:

1. perf and latencytop

sudo perf record -e sched_switch -a  
sudo latencytop

2. /proc Interface

Check RT parameters:

cat /proc/sys/kernel/sched_rt_period_us  
cat /proc/sys/kernel/sched_rt_runtime_us

Key takeaways

  • Real-time scheduling requires PREEMPT_RT kernel patches and careful configuration.
  • Use SCHED_FIFO for hard deadlines and SCHED_DEADLINE for advanced workflows.
  • Optimize latency by adjusting sched_rt_period_us/sched_rt_runtime_us and binding tasks to CPUs.
  • Monitor with perf and latencytop to identify bottlenecks.
  • Prioritize critical tasks over non-RT workloads to ensure deterministic behavior.