reliability
Imagine a car that starts every single morning for ten years without ever leaving you stranded. You would call it reliable — not because it is fast, but because it keeps working correctly for a long stretch without breaking. Reliability in computer hardware is exactly this idea: the property of a system continuing to do the right thing, without failure, over a period of time. A laptop that crashes once a year is more reliable than one that crashes once a week, even if both have the same speed.
Engineers make this precise by talking about a system that is 'up' (working) and the chance it stays up. Reliability is often written as a probability that the system survives without failure from time 0 up to some time t. A common simple model assumes failures arrive at a roughly constant rate, so reliability decays exponentially with time — the longer you run, the more likely some part has failed by now. From that rate engineers derive the headline number most people quote, the mean time between failures (MTBF): the average length of a working stretch (see mean-time-between-failures). Higher MTBF means higher reliability.
A crucial and honest point: reliability is about correct, uninterrupted service, which is not the same as availability. Availability asks 'what fraction of the time is it up?', and a system can be highly available yet not very reliable if it fails often but is repaired almost instantly each time (see availability). Reliability also is not free: it is bought with redundancy, error-correcting codes, and careful design, all of which cost money, power, and chip area. The right level of reliability is a deliberate engineering choice, much higher for a bank's database or a spacecraft than for a game console.
A disk drive rated for an MTBF of 1.2 million hours does not mean each disk lasts 1.2 million hours (that is 137 years). It means that across a large fleet, in any given hour you expect about one failure per 1.2 million running drives. Run 1,200 drives and you expect roughly one failure every 1,000 hours — about once every six weeks.
MTBF describes a population's failure rate, not the guaranteed lifetime of one unit — a common misreading.
Reliability and availability are different. A system that fails constantly but is repaired in milliseconds can be highly available yet poorly reliable; choose the metric that matches what users actually feel.