The 100% Myth
Imagine your boss asks, "How reliable is our new service?" You might look at a dashboard and say, "The CPU is at 40%, so it's fine." But CPU utilization tells you nothing about whether users are successfully completing their tasks. A service can have idle CPUs and still be returning errors to every single user.
Your next instinct might be to say, "We need 100% uptime. It should never fail." This is a trap.
Achieving 100% reliability is practically impossible and astronomically expensive. Every extra "nine" of reliability (going from 99% to 99.9% to 99.99%) costs exponentially more in engineering time and infrastructure. More importantly, chasing 100% means you can never take risks, never push new features, and never run experiments — because any change might cause a momentary failure. You end up with a frozen, stagnant system that is technically "reliable" but commercially useless.
The Cost of Each Nine
| Availability | Downtime per Year | Downtime per Month |
|---|---|---|
| 99% ("two nines") | ~87.6 hours | ~7.3 hours |
| 99.9% ("three nines") | ~8.76 hours | ~43.8 minutes |
| 99.99% ("four nines") | ~52.6 minutes | ~4.4 minutes |
| 99.999% ("five nines") | ~5.26 minutes | ~26 seconds |
The jump from four nines to five nines requires reducing your allowed downtime from 52 minutes to 5 minutes per year. That demands redundant everything, zero-downtime deployments, automated failover, and a dedicated team to maintain it. For most services, that cost is not justified by the business value.
Engineering for User Happiness
Site Reliability Engineering (SRE), a discipline pioneered by Google, introduced a framework to solve this. Instead of aiming for impossible perfection, we aim for "reliable enough to keep users happy."
This framework uses precise measurements and agreed-upon goals to balance the need for reliability with the need for feature velocity. It's built on four key concepts: SLIs, SLOs, SLAs, and Error Budgets. Each concept builds on the previous one, forming a complete system for making data-driven decisions about risk.