Storage
Storage is the part of a computer system that retains data and programs for later use, keeping them available after the program that created them has ended and, in most forms, after the power is switched off. It is the fourth basic operation, the one that gives a computer a lasting memory, and it is distinct from the main memory a running program uses as a temporary workspace.
The defining property of storage is persistence at a trade-off against speed. Working memory is fast but loses its contents when power goes; storage is slower, much larger, cheaper per unit of capacity, and non-volatile. This produces a hierarchy: a few kilobytes of registers and cache close to the processor, gigabytes of main memory, hundreds of gigabytes to terabytes of local storage, and effectively unlimited remote storage over a network, with each step down being larger, cheaper, and slower. Software and operating systems are built around that hierarchy, moving data upward before use and writing results back down when they must be kept.
Storage technology has moved through several physical principles. Punched cards and paper tape held data as holes. Magnetic recording took over: IBM's 305 RAMAC of 1956 introduced the hard disk drive, storing about 5 megabytes on a stack of large platters, and modern hard drives pack terabytes onto platters read by heads that float a few nanometers above the surface. Optical discs such as the compact disc, introduced in 1982, store data as microscopic pits read by a laser. Flash memory, invented by Fujio Masuoka at Toshiba in the 1980s, stores charge in transistors with no moving parts and is the basis of solid-state drives, phone storage, and USB sticks. Magnetic tape remains in use for archives because of its low cost per byte.
Storage is organized in layers of abstraction above the raw hardware. A drive exposes numbered blocks of bytes; a file system on top of those blocks gives them names, folders, and permissions, so that users and programs deal with files rather than block numbers. Databases add structure and fast lookup on top of files. Redundancy is added for reliability: RAID arrays spread data across several drives so that one failure does not lose it, and a backup keeps a separate copy against deletion, corruption, and ransomware. Storage is also placed across a network: network-attached storage on a local network and cloud object storage, offered at scale by Amazon S3 from 2006, let data live far from the machine using it.
Capacity, speed, and durability are three different properties, and a common error is to treat them as one. A larger drive is not a faster one; an SSD may offer far lower latency than a hard drive while costing more per terabyte. Durability is the other pitfall: storage media degrade, flash cells wear after many writes, hard drives fail mechanically, and file formats and readers become obsolete, so data that is merely kept is not necessarily data that can be read in twenty years. Deleting a file generally removes its directory entry rather than erasing the bytes, which is why secure disposal requires overwriting or encryption. For machine learning, storage speed often bounds training: feeding large datasets to accelerators fast enough is a design problem in its own right.
Another frequent confusion is between storage and memory, which are both called memory in everyday speech. A phone advertised with 8 gigabytes of memory and 256 gigabytes of storage has two different things: the first is the temporary workspace, the second is where apps and photos persist. Adding storage does not speed up a machine that is short of memory, and adding memory does not create room for more files.
Read storage as the computer's long-term memory — slower and larger than working memory, but persistent — whose real value depends on durability, redundancy, and the discipline of keeping more than one copy.