CPUaffinitet

CPU affinity, also known as processor affinity or CPU pinning, is a scheduling feature that binds a process or thread to a specific subset of CPUs. In Danish, the term is CPUaffinitet. The binding is implemented by a mask or list of CPUs that the task is allowed to run on. If a task has affinity set, the system scheduler will not migrate it to CPUs outside the allowed set (except in some fault or emergency situations).

Purpose and effects: Affinity can improve performance by increasing cache locality and reducing cross-core traffic, as a thread repeatedly executes on the same cores. It can also help meet real-time or latency requirements by limiting contention on busy systems. However, overly strict affinity can degrade throughput by preventing the scheduler from balancing load across cores.

Implementation: Operating systems expose different interfaces. Linux provides sched_setaffinity and sched_getaffinity, and the command line tool taskset can adjust a process’s CPU mask. Windows offers SetProcessAffinityMask and SetThreadAffinityMask to pin processes or threads. macOS provides more limited and heuristic-based affinity controls, focusing on affinity tags rather than strict pinning in most cases.

Typical use cases: high-performance computing, database servers, media encoding, and real-time applications are common scenarios for manually setting affinity to optimize CPU cache reuse and predictability.

Caveats: Affinity policies can become complex to manage, may hinder dynamic load balancing, and reduce portability across systems. Changes should be tested to ensure they yield measurable benefits in the target workload.