Cloud Platform
Uptime percentages are quoted constantly and examined rarely. They're worth examining, because the gap between what a number sounds like and what it contractually guarantees is where unpleasant conversations live.
The numbers, converted into time
Availability percentages only become meaningful as durations. Per year, the allowed downtime is roughly:
- 99% — about 3 days and 15 hours
- 99.9% — about 8 hours and 45 minutes
- 99.95% — about 4 hours and 22 minutes
- 99.99% — about 52 minutes
- 99.999% — about 5 minutes
Three nines — the most commonly advertised figure — permits a full working day of outage per year and remains fully compliant. Whether that's acceptable depends entirely on your business, which is exactly why the number alone tells you so little.
The three questions that matter more than the percentage
1. What counts as downtime?
This is defined in the agreement and it is rarely what a user would say. Common carve-outs: scheduled maintenance windows, degraded-but-responding performance, failures attributed to your own configuration, network problems outside the provider's boundary, and force majeure.
A service that is technically responding while too slow to use is, under many agreements, available. Read that definition first; it constrains everything else.
2. What is the measurement window?
Availability measured monthly and availability measured annually are very different promises. Under a monthly window, 99.9% allows about 43 minutes per month — and a single 43-minute outage each month keeps the provider compliant all year while delivering over eight hours of annual downtime.
3. What is the remedy, and who has to claim it?
Almost universally, the remedy is a service credit — a percentage off a future invoice. It is not compensation for your losses, and it is usually capped at a fraction of the monthly fee.
More importantly: in most agreements you must claim it, within a defined period, with your own evidence. Credits that are never claimed are never paid, which is a significant part of why they're offered.
The measurement question nobody asks: who monitors compliance, and from where? If availability is measured by the provider's own internal monitoring, from inside their own network, that measurement can genuinely differ from your users' experience. Independent external monitoring is worth having even when you trust the provider entirely — it makes the conversation factual.
What actually produces uptime
The percentage is an outcome, not a mechanism. What produces it:
- No single point of failure in power, network, storage, or compute — and evidence that failover has been tested rather than merely designed
- Change management, because most outages are caused by changes, not by hardware
- Monitoring that alerts a human who has the access and authority to fix it, at the hour it happens
- Capacity headroom, because saturation looks exactly like an outage to a user
A provider who can describe their failover testing regime in specific terms is telling you more about your real availability than any figure on a pricing page.
What to ask before signing
- How is downtime defined, and does degraded performance count?
- Is availability measured monthly or annually?
- Is scheduled maintenance excluded, and how much notice do we get?
- Who measures it, from where, and can we see the data?
- What's the remedy, is it automatic, and what's the claim deadline?
- When did you last test failover, and what did you find?
The last question is the most revealing. A provider who has recently tested failover and found something is a provider who tests failover.
