Availability and reliability - the metrics that matter

Availability is the percentage of time a service is working. Reliability is the probability that a service works at a given time. This lesson covers the metrics and their trade-offs.

The 9s

Each “9” requires 10x more effort than the last. Going from 99.9% to 99.99% is much harder than 99% to 99.9%.

MTBF and MTTR

• MTBF (Mean Time Between Failures): average time between failures.
• MTTR (Mean Time To Repair): average time to restore service.

Availability = MTBF / (MTBF + MTTR)

To improve availability:

• Increase MTBF: prevent failures (better hardware, better software, monitoring).
• Decrease MTTR: faster detection (monitoring), faster recovery (runbooks, automation).

Often decreasing MTTR is cheaper than increasing MTBF. A fast restore makes 99.9% achievable; preventing the failure in the first place is much harder.

SLA

An SLA (Service Level Agreement) is a contract with the customer. It typically specifies:

• Availability target (e.g. 99.9%).
• Response time (e.g. p99 < 1 second).
• Support response time (e.g. 1 hour for critical).
• Credits for missing the SLA.

SLAs are a business commitment. Designing to meet the SLA is an engineering discipline.

Per-service targets

Different services have different availability targets:

• Authentication: 99.99% (downtime is felt fleet-wide).
• Customer-facing API: 99.9%.
• Internal tooling: 99% (downtime is acceptable).
• Background processing: best-effort.

Match the engineering to the target. Over-engineering a 99% service for 99.99% wastes resources.

Knowledge check