Availability — the nines and their cost
What 99.9% vs 99.99% actually means, and what each nine costs.
Availability is a promise — the fraction of requests that succeed (or the fraction of time your system responds) over a measurement window. Every additional 9 costs roughly 10× more engineering + infrastructure. Pick the smallest number that keeps your business alive.
99.99% availability — what does it mean?
Infra required: Multi-AZ, auto-failover, on-call rotation
Cost: 5-10× baseline (~$5-20K/mo)
The senior insight — availability math is multiplicative
If a request touches 5 services each at 99.9% availability, total availability is 0.9995 = 99.5%. The path availability degrades. This is why microservice architectures need higher per-service availability than monoliths to feel “the same.” And why isolation via async / caching / graceful degradation is the actual scaling story.
The 4 rules of setting an SLO
- Set SLO below your actual capability. Header room = flexibility for deploys + chaos.
- Measure from the user's perspective (browser telemetry), not the server's (5xx rate).
- SLO tied to error budget. If you exceed budget, freeze features + focus on reliability.
- Different SLOs per endpoint. Payments = 99.99%. Recommendations = 99.5%. Get creative.
Practice what you just read
Every foundation concept has a companion quiz to close the loop.