Apple has open-sourced its new Clutch Scheduler for the XNU kernel, designed to improve thread scheduling across its platforms. The scheduler addresses limitations in the traditional Mach scheduler by implementing a hierarchical approach that schedules groups of threads rather than individual threads. This design aims to better handle the diverse requirements of modern computing, from latency-sensitive UI interactions to background batch workloads like photo synchronization and source compilation.

The Clutch Scheduler operates at three levels: scheduling buckets, thread groups, and individual threads. The highest level uses an Earliest Deadline First (EDF) algorithm to prioritize different Quality of Service (QoS) classes, ensuring low-latency access for high-priority tasks while preventing starvation of lower-priority operations. Each scheduling bucket has a predefined Worst Case Execution Latency (WCEL) value that determines when it should be scheduled next, mimicking the decay curve of the traditional Mach timesharing algorithm.

To address bursty workloads and maintain responsiveness, the scheduler implements a warp mechanism that gives higher-priority buckets a bounded advantage when they become runnable. This prevents situations where high-priority tasks get stuck behind lower-priority ones in heavily loaded systems. The design also tackles issues like inaccurate CPU accounting and poor isolation between latency-sensitive and non-latency-sensitive threads. By grouping threads and making decisions at multiple levels, the Clutch Scheduler provides better reasoning about user workloads as a whole rather than treating each thread in isolation.