Writeback

Writeback is a caching and data-management technique in which modifications to data are written to a fast cache or buffer first and are written to the backing storage later. The main goal is to improve write performance and reduce contention on slower storage by batching writes and keeping frequently changed data in fast memory. Writeback is contrasted with write-through, where every write to the cache is immediately and synchronously written to storage.

In hardware caches, a write-back cache stores modified blocks as dirty entries. A block is marked dirty when written, and the actual write to main memory or to disk occurs when the block is evicted, when a flush is triggered, or at scheduled intervals. This reduces write bandwidth and memory traffic and can improve overall throughput, but it increases the risk of data loss if power is lost before the cache is flushed. The risk is mitigated by using non-volatile memory or power-loss protection such as battery-backed caches.

Writeback is also employed in storage controllers, operating systems, and certain database systems to optimize I/O performance. In these contexts, a cache may defer writes to the underlying storage and consolidate multiple updates. While this can boost latency and throughput, it requires careful handling of data durability and cache coherence to avoid inconsistencies during failures or crashes.

Advantages of writeback include higher write throughput, lower latency for individual writes, and reduced wear on backing storage. Disadvantages include potential data loss on power failure, increased complexity for ensuring data integrity, and more complex recovery procedures after crashes. Related concepts include write-through caching and asynchronous write-behind strategies.