Uptime is the time a system, application or service is up and running, often expressed as a percentage of a month or year. Strictly, it describes whether something is running. When a provider quotes “uptime” in a service level agreement, it usually means something narrower and contractual: service availability as measured by that contract’s own formula, definitions and exclusions. A 99.9% uptime SLA therefore commits to whatever that formula says, which may not match the outages your staff actually experienced, so read the formula before comparing numbers.
At a glance
- Uptime is the time something is running; a provider’s “uptime SLA” typically measures contractual service availability as a percentage, using its own formula, definitions and exclusions.
- Each additional “nine” cuts allowed downtime by a factor of ten: 99.9% is about 8.8 hours per year, 99.99% about 53 minutes.
- Missing an uptime target usually earns service credits, not compensation for business losses.
- Uptime is an outcome; high availability design and fast recovery are how you get it.
What problem it solves
Buyers need a common way to compare reliability and hold providers to it. Uptime gives a single number for that: it sets an expectation in the contract, a threshold for credits, and a target for internal teams designing systems. It also forces a useful internal question: how much downtime can each system tolerate, and what is it worth paying to reduce it?
The trap is treating the percentage as the whole answer. Two services with the same 99.9% target can deliver very different experiences depending on whether outages are many short blips or one long event, whether maintenance is excluded, and whether “available” means usable by your staff or just reachable by the provider’s monitoring.
How it works
The calculation. A typical SLA formula is (eligible measurement time − counted downtime) ÷ eligible measurement time × 100. Contracts define each part differently: which minutes are eligible (for example, whether maintenance windows are removed from the denominator), which events count as downtime in the numerator, and what is excluded altogether. Many commercial SLAs measure per calendar month, so the allowance resets every month.
What counts as down. The contract defines it. For an internet circuit it might be total loss of service; for a cloud platform it might be failed requests above a threshold. Partial degradation, such as heavy packet loss or slow responses, often doesn’t count unless the SLA has separate performance terms.
Exclusions. Planned maintenance, customer-caused problems, third-party failures and force majeure are commonly excluded. Some providers require you to open a ticket for the outage clock to start.
Measurement and credits. The provider usually measures uptime with its own monitoring. If the target is missed, you request credits on a published scale, often within a claim window.
Five nines and other levels
Allowed downtime at common targets, by simple arithmetic before any contractual exclusions (annual figures assume a 365-day year; monthly figures a 30-day month):
- 99.5%: about 43.8 hours per year, 3.6 hours per month.
- 99.9% (three nines): about 8.8 hours per year, 43 minutes per month.
- 99.95%: about 4.4 hours per year, 22 minutes per month.
- 99.99% (four nines): about 53 minutes per year, 4.3 minutes per month.
- 99.999% (five nines): about 5.3 minutes per year, 26 seconds per month.
Five nines is a common goal for voice and critical infrastructure, but it usually describes the design of a whole system with redundancy and automatic failover, not something one circuit or server delivers on its own.
When it matters for buyers
- When signing for internet, cloud or colocation. Compare uptime definitions, exclusions and credits, not only percentages. Our dedicated internet access overview covers what circuit SLAs typically include.
- After an outage. Check whether the event counted against the SLA and claim credits within the window.
- When setting internal targets. Decide which systems need which level, then design for it rather than buying a number.
- When stacking services. End-to-end uptime is lower than each component’s: two services at 99.9% that both have to work give roughly 99.8% together unless one can cover for the other.
Questions to ask vendors
- How do you define “available,” and does degraded performance count as downtime?
- Is uptime measured monthly or annually, and from where?
- What is excluded, and when and how often can planned maintenance happen?
- What credits apply at each level, how do we claim them, and is there a cap?
- What was your actual uptime for this service over the last 12 months?
- Where are the single points of failure between your network and our site?
How it differs from high availability (HA)
Uptime is a measured result: how much of the time a system or component was running. Availability, which measures whether the service was actually usable, is what provider “uptime” SLAs commonly calculate. High availability (HA) is the design approach used to raise it, such as redundant components, failover and removing each single point of failure (SPOF). A service level agreement (SLA) is the contract that states the availability target, how it is measured and the remedy. Mean time to recovery (MTTR) describes how fast service comes back after a failure, which, along with how often failures happen, shapes the uptime and availability you actually get.