readahead

Readahead, often written read-ahead, is a prefetching technique used by operating systems and some applications to load data into memory before it is actually requested. The goal is to hide the latency of slower storage devices by presenting already cached data to the consumer when the read occurs.

In many Unix-like systems, readahead is implemented in the block layer and the filesystem code. When a read request is detected to be part of a sequential access pattern, the system may issue additional read requests for following blocks so that the page cache contains the next data blocks by the time the program reads them. The technique relies on heuristics to detect sequential or predictable access and to decide how much data to prefetch.

Tuning and scope: The amount of data prefetched is configurable. On Linux, parameters such as read_ahead_kb (for devices) control the amount of data brought into the cache. Too large a value can waste I/O bandwidth or evict useful data; too small may miss opportunities to hide latency. Readahead is typically disabled or reduced for random I/O patterns and for devices with high latency or limited bandwidth.

Impact and usage: Readahead can improve throughput in workloads with predictable sequential I/O, such as reading large files or streaming data, while being less helpful for random access patterns. It works in concert with the page cache and I/O schedulers to balance memory usage and latency.

See also: Read-ahead, page cache, I/O scheduler, prefetching.