Definition
Prefetching is a performance optimization technique used in computing whereby data, instructions, or resources are retrieved and loaded into a faster-access storage medium before they are actually required for processing. The aim is to reduce latency and improve overall system throughput.
Overview
Prefetching is employed in various layers of computer architecture and software engineering, including processor caches, disk I/O subsystems, web browsers, and database management systems. By anticipating future memory accesses or I/O requests, systems can overlap the latency of data retrieval with useful work, thereby improving execution speed. Different prefetching strategies may be deterministic—based on static analysis or predetermined patterns—or dynamic—relying on runtime heuristics and machine learning models to predict future accesses.
Etymology/Origin
The term combines the prefix “pre‑,” meaning “before,” with the verb “fetch,” meaning “to retrieve.” It entered technical literature in the late 1970s and early 1980s, coinciding with research on cache memory hierarchies and the need to mitigate the growing performance gap between CPUs and main memory.
Characteristics
| Characteristic | Description |
|---|---|
| Scope | Can be applied at multiple levels: CPU instruction prefetch, data cache prefetch, storage prefetch, network/web prefetch, and application‑level prefetch. |
| Trigger Mechanism | - Static: compiler directives, predetermined access patterns. - Dynamic: hardware branch predictors, software profiling, adaptive algorithms. |
| Temporal Locality Exploitation | Prefetching relies heavily on the principle that recently accessed data is likely to be accessed again soon. |
| Spatial Locality Exploitation | Many schemes prefetch adjacent blocks of memory assuming sequential access. |
| Cost/Benefit Balance | Effective prefetching reduces stall cycles, but incorrect predictions can waste bandwidth, increase cache pollution, or cause unnecessary I/O. |
| Implementation | Implemented in hardware (e.g., hardware prefetch engines in modern CPUs), firmware (e.g., disk controllers), operating system kernels, and application software (e.g., lazy loading in web browsers). |
| Metrics | Success rate (percentage of useful prefetches), coverage (ratio of needed data prefetched), latency reduction, and overhead (extra bandwidth or power consumption). |
Related Topics
- Cache Memory – Hierarchical storage where prefetching often targets upper‑level caches.
- Branch Prediction – Predicts program flow to aid instruction prefetch.
- Speculative Execution – Executes instructions before it is known they will be needed, sometimes combined with prefetching.
- Read‑Ahead – A specific form of disk‑level prefetch that reads sequential blocks ahead of the current request.
- Lazy Loading – Defers loading of resources until they are actually needed; the opposite approach to prefetching.
- Prefetch Abortion – Mechanisms to cancel or ignore prefetches that become unnecessary due to changed program behavior.
Prefetching remains a critical component in modern high‑performance computing, influencing the design of processors, storage devices, and networked applications.