Secondary storage

RAM forgets everything when the power goes, so a computer needs somewhere permanent for its operating system, programs and files: secondary storage. It is non-volatile, much bigger and much cheaper per gigabyte than RAM, and much slower, so programs are copied from it into RAM before they run. Internal storage is built into the device; external storage plugs in or is reached over a network.

A hard disk drive stores data as tiny magnetised areas on metal platters spinning thousands of times a minute. A read/write head on an arm floats just above the surface. To read a file the arm must move to the right track (seek time) and then wait for the right part of the platter to spin round under it (rotational delay). Those mechanical delays make HDDs slow, especially for many small scattered files, but they store a lot of data cheaply.

A solid state drive stores data in NAND flash memory chips, managed by a controller chip, with nothing moving. Any location can be read about as fast as any other, so SSDs are much faster, as well as silent, lighter and more robust. Small M.2 SSDs plug straight into the motherboard; the fastest (NVMe) connect directly to the CPU's fast lanes. eMMC is also flash, but simpler and soldered onto the board: cheaper and slower, found in budget phones, tablets and laptops.

Flash (in SSDs, eMMC, USB drives and memory cards) stores each bit as electric charge trapped in a tiny cell, which stays put with the power off. The cells are grouped into pages (a few kilobytes), and pages into blocks. Data can be written a page at a time, but a cell cannot simply be overwritten: it must be erased first, and erasing works only on a whole block at once. So changing one page means the controller writes the new version somewhere already empty and tidies up the old block later.

Every erase wears a cell a little, and after a limited number of write/erase cycles it can no longer hold its charge reliably. The controller therefore uses wear levelling: it spreads writes evenly over all the blocks, so no block is worn out by a file that changes constantly, and it keeps spare blocks to swap in for any that fail. Cheaper flash stores more bits per cell (one bit is SLC, two MLC, three TLC, four QLC): denser and cheaper per gigabyte, but slower and able to survive fewer writes. For everyday use a modern SSD lasts for years; it matters for drives written to all day, such as busy servers.

• External HDDs and SSDs work just like the internal kinds, in a case with a USB cable: an SSD for fast, knock-proof portability; an HDD for cheap, large backups.
• USB flash drives and memory cards (SD, microSD) are small flash devices. Cards are made for cameras, phones and consoles: tiny, tough and removable, though slower than an SSD.
• Optical discs (CD, DVD, Blu-ray) are read by a laser. Very cheap to produce in bulk and fine for distribution or archives, but slow, easily scratched, and many devices no longer have a drive.
• Network attached storage is a box of drives on the network that many people can share. It usually combines its drives using RAID, so the data survives if one drive fails, and more drives can be added as it fills up.

Storage on someone else's servers, reached over the internet, is cloud storage: see A1.1.9.

CD, DVD and Blu-ray work the same way. Each newer format uses a laser with a shorter wavelength (Blu-ray's is blue-violet, hence the name), which can read smaller pits packed closer together, so far more fits on the same size of disc.