CONCEPTS
Availability
The proportion of time a system is operational and accessible.
Availability = Uptime / (Uptime + Downtime)
The proportion of time a system is operational and accessible.
The Quest for "Nines"
Availability is usually expressed as a percentage of time that a service is working. "The Nines" refer to the number of 9s in that percentage.
The Cost of Nines
Each additional "9" costs exponentially more money to achieve.
- 99% (Two Nines): Good for internal tools. Down 3.6 days/year. (Cheap).
- 99.9% (Three Nines): Industry Standard. Down 8.8 hours/year. (Affordable).
- 99.99% (Four Nines): High Availability. Down 52 mins/year. (Expensive).
- 99.999% (Five Nines): Telco/Medical Grade. Down 5 mins/year. (Extremely Expensive).
Availability vs. Reliability
- A car that breaks down every day but is fixed in 1 minute has high Availability (it runs most of the time) but low Reliability (it breaks often).
- A car that never breaks down is both.
ExThe Region Outage
"A company hosted their app in a single AWS region (us-east-1). The region went down for 4 hours."
Impact
Availability dropped to 99.5% for the year, violating customer contracts.
Resolution
They moved to a Multi-Region Active-Active architecture to achieve 99.99% availability, costing 2x more in infrastructure.
Why Availability Matters
Availability directly impacts revenue and user experience. Every minute of downtime costs money.
High availability (99.9%+) requires redundancy and careful architecture.
Common Pitfalls
Chasing Five Nines
Aiming for 99.999% unavailability when your ISP only gives you 99.9%. You cannot be more available than your dependencies.
Industry Benchmarks
Five Nines< 5 min/year
99.999%
Four Nines< 53 min/year
99.99%
Three Nines< 8.8 hours/year
99.9%
Two Nines< 3.7 days/year
99%
Related Terms
Frequently Asked Questions
Is 99.9% good enough?
For most SaaS products, yes. Users generally tolerate ~45 minutes of maintenance per month.
Does scheduled maintenance count as downtime?
It depends on your user contract (SLA). Usually, yes, because the user cannot use the service.