99.9% Availability

The Downtime Math

 99.9% → 0.1% failure budget

 per YEAR    ≈ 8.77 hours down     (525,600 × 0.1%)
 per MONTH   ≈ 43.8 minutes        (43,800 × 0.1%)
 per WEEK    ≈ 10.1 minutes
 per DAY     ≈ 1.44 minutes

Eight-plus hours a year sounds loose until you note the monthly shape: under 44 minutes per month covers every deploy incident, dependency hiccup, and human error combined.

What Three Nines Demands

The step from 99% to 99.9% eliminates single-machine thinking:

Requirement	Why
Multi-AZ within region	One AZ’s outage must not be user-visible
Automated failover (DB primary)	Manual promotion takes longer than the budget
Health checks + auto-removal	Dead nodes can’t receive traffic for minutes
Zero-downtime deploys	Rolling restarts can’t consume the whole month
Monitoring with alerting	Undetected partial outages burn budget invisibly

Notice what is not yet required: multi-region, active-active, exotic consistency machinery. Three nines is achievable inside one cloud region with disciplined basics.

The Budget in Operational Terms

 monthly allowance: ~43 minutes

 typical consumers of the budget:
 routine rolling deploys         ~minutes of elevated error rate each
 one AZ blip per quarter         absorbed by design ✓
 a bad config push               15–30 min if rollback is practiced
 dependency outage (PSP, SMS)    partially yours to absorb
 
 teams that don't rehearse rollback routinely spend their entire
 budget on ONE incident — then operate "degraded" for weeks

This is why error-budget discipline (SRE practice) pairs naturally with three nines: the number becomes a management tool, not just a badge.

Who Targets Three Nines

• Consumer apps at scale: feeds, commerce browsing, social — staleness and brief errors tolerated; unavailability is not.
• Standard SaaS products: contractual norms cluster here.
• Internal platforms serving many teams: cascading blast radius justifies the tier above two nines.

Payments capture, trading, emergency systems look past this tier entirely.

The Cost Curve Step

Each additional nine multiplies infrastructure cost roughly 2–10x depending on starting posture:

 99%    : single-AZ, simple            baseline $
 99.9%  : multi-AZ + automation         ~1.5–2x      ← biggest jump in CAPABILITY
 99.99% : multi-region-ready, more       multiples more

Three nines is widely considered the knee of the curve — the last tier where added nines buy disproportionate capability per dollar.

Interview Framing

“99.9%” is the correct default answer for most consumer designs with justification attached: name the multi-AZ posture and zero-downtime deploys it implies. Strong candidates immediately split exceptions (“payment capture gets its own stricter path”). The follow-up probe is always “what does that mean architecturally?” — have the requirements table ready.