Uptime
MTTF
Also called mean time to failure.
MTTF is the average time a non-repairable component runs before failing. It describes the part, not the response.
How it is measured
Divide total operating hours across a group of units by the number of failures. 40 disks running 1,000 hours with 2 failures gives 20,000 hours. It says nothing about any single unit, and it assumes a fairly constant failure rate.
For web operations it applies to parts: disks, power supplies, certificates, VPS instances. It informs spares and refresh cycles. It is not a promise about any one machine.
Worked example
A hosting reseller tracks 120 SSDs over a year, about 1.05 million disk-hours. 5 die. MTTF is roughly 210,000 hours, about 24 years, which sounds absurd for a device with a 5-year warranty.
The figure is a population rate during useful life, not an expected lifespan. The practical reading is that about 4 percent of the fleet fails each year, so keep spares and replace at the warranty mark.
How it differs
MTTF is time until failure for something you replace. MTTR is time to recover after failure. MTTF excludes repair. MTTR excludes how often it breaks. In simple models, availability is MTTF divided by MTTF plus MTTR.
Common errors
Reading it as lifespan. Using it on a single server. Mixing it up with MTBF. Applying it to software releases, whose failure pattern changes with every deploy. Ignoring wear-out near end of life.
In practice
Use it for hardware fleets and renewal schedules: count failures per quarter and compare. For a single site, track your own incident frequency instead. If you run hardware, keep spares for the parts that fail most.