MTTR, MTTF and the Availability Equation

Your SLO measures availability. That availability is the result of two opposing forces: how often your system fails, and how fast you fix it. Acting on either one has the same mathematical effect — but very different costs.

Definitions

MTTF (Mean Time To Failure): average operating time between two failures. The higher it is, the more reliable your system. MTTR (Mean Time To Recovery): average time from failure onset to return to normal. It includes detection, diagnosis, correction, and verification. MTBF (Mean Time Between Failures) = MTTF + MTTR.

The availability formula

Availability is calculated directly from MTTF and MTTR. It's the ratio of "healthy" time to total time. Example: if your service fails on average every 15 days (MTTF = 21,600 min) and you repair it in 21.6 minutes on average, you achieve exactly 99.9% availability — meaning the full error budget of a 99.9% SLO is consumed.

The MTTR lever is often more effective

Halving MTTR has exactly the same effect on availability as doubling MTTF. Mathematically, M/(M+T/2) = 2M/(2M+T). But improving MTTF means reducing failure frequency — which often requires infrastructure redesign, more testing, chaos engineering. Improving MTTR means writing better runbooks, automating rollbacks, improving dashboards, and training on-call. This is generally faster and cheaper.

Detection time: the hidden part of MTTR

Most teams measure MTTR from alert acknowledgement, not from the actual start of the failure. MTTD (Mean Time To Detect) — the time between failure and first alert — is often invisible in post-mortems. A service that fails at 2am and isn't detected until 8am has a real MTTR of 6h+ even if resolution took 20 minutes. Your SLOs measure real availability, not availability from the alert.