ReadCopyUpdate

Read-Copy-Update, commonly abbreviated as RCU, is a synchronization mechanism used in concurrent systems to optimize read-mostly data access. It enables multiple readers to access shared data without taking locks, while writers update the data by creating a new version and then switching pointers to the new version. The old version remains accessible to readers until it is guaranteed that no reader can still access it, at which point the memory can be reclaimed. This guarantee is provided by a grace period and callbacks that reclaim memory after that period.

In typical RCU operation, readers access data in a read-side critical section that is lightweight and non-blocking. Writers perform updates by duplicating the affected data, applying changes to the new copy, and atomically swapping the pointer to the new version. After the swap, the writer schedules reclamation of the old version and waits for a grace period during which all pre-existing readers finish. Once the grace period expires, the old data can be freed. Several variants use different strategies to detect quiescent states or to manage grace periods, such as Quiescent-State-Based Reclamation (QSBR) and various RCU callback mechanisms.

RCU is widely used in operating system kernels and other high-performance systems where read operations greatly outnumber writes. Its main advantage is near-elimination of read-side contention, allowing lock-free reads. Its drawbacks include increased writer complexity, potential latency in reclaiming old data, and more intricate memory ordering and lifecycle considerations. RCU is most effective for data structures that are read-intensive and can tolerate deferred reclamation.