Prefetching

Prefetching is a performance optimization in which data or instructions are moved to a faster storage layer before they are needed, with the goal of hiding memory latency and increasing system throughput. Prefetching can occur at multiple levels, from processor caches to storage subsystems and even networked resources.

In CPUs, prefetching can be hardware-driven, where the memory subsystem detects access patterns and issues prefetch requests ahead of demand, or software-driven, where compilers or programmers insert prefetch hints or use special intrinsics. Hardware prefetchers typically target regular, stride-based access, while software prefetching can handle irregular patterns but risks misprediction and cache pollution if mis-tuned.

Operating systems and storage subsystems implement prefetching through read-ahead or prefetch algorithms that fetch data blocks before the application requests them, improving performance for sequential or predictable workloads. Disk and memory prefetchers rely on historical access patterns and can adapt to changing workloads, trading off bandwidth against latency.

In web environments, browsers may prefetch or prerender resources associated with a page, or DNS prefetching to reduce connection setup delays. While these techniques can improve perceived responsiveness, they increase bandwidth consumption and may raise privacy concerns by accessing resources users have not yet requested.

Effectiveness depends on workload characteristics, data locality, and system resources. Poorly tuned prefetching can evict useful data or saturate bandwidth, while well-tuned prefetchers can yield significant latency reductions and smoother performance.

See also caching and speculative execution.