A service level agreement (SLA) is an agreement in which a service provider commits to a level of service, such as uptime, and defines what happens if it fails to meet it. In cloud services, SLAs are expressed as numbers like “99.9% monthly uptime,” and when the provider falls short they usually give a service credit that reduces future bills. This article explains how to read uptime percentages, what to check in a cloud provider’s SLA, and how to set an SLA for a service you provide.
SLA, SLO and SLI
Three similar terms are worth separating:
- SLI (service level indicator): the metric you actually measure, such as the share of successful responses or the share of responses served within one second
- SLO (service level objective): the value the provider aims for on that indicator, for example 99.95% successful responses per month
- SLA (service level agreement): the subset of objectives promised to customers, together with what happens when they are missed, such as service credits
Keep your internal SLOs stricter than the SLA you promise customers. If you start acting as soon as an SLO is breached, you can fix things before the SLA itself is at risk.
How much downtime an uptime percentage allows
Uptime is “time the service was available ÷ time covered.” For a 30-day month (43,200 minutes), the allowed downtime works out as follows:
| Monthly uptime | Allowed downtime (30-day month) | Per year |
|---|---|---|
| 99% | about 7 h 12 min | about 3 days 15 h |
| 99.5% | about 3 h 36 min | about 1 day 20 h |
| 99.9% | about 43 min | about 8 h 46 min |
| 99.95% | about 22 min | about 4 h 23 min |
| 99.99% | about 4 min | about 53 min |
Each extra nine cuts the allowed downtime by a factor of ten. At 99.99%, there is no time for a person to notice and fix a problem by hand; you need redundancy across locations and automatic failover.
Examples of published cloud SLAs
Here are some published SLAs (as of November 2026):
- Google Cloud Compute Engine: instances spread across multiple zones target at least 99.99% monthly uptime, and a single instance at least 99.9% (99.95% for memory-optimized machine types). When the target is missed, a percentage of the monthly bill is credited to future bills. The SLA states that this is the customer’s sole and exclusive remedy
- Google Workspace: covered services are committed to 99.9% uptime in any calendar month
Note that the same provider commits to different numbers depending on your architecture, a single instance or several zones. If you use a provider’s SLA to set your own service level, read the SLA for the plan and architecture you actually use.
Six things to check in an SLA
- What counts as downtime. Only complete unavailability, or also periods where the error rate exceeds a threshold? Check the measurement interval too, every minute or every five minutes
- Scheduled maintenance. Whether announced maintenance windows are excluded from the uptime calculation
- Exclusions. Problems caused by the customer’s own configuration, causes beyond the provider’s reasonable control such as major disasters, and pre-GA features are typically excluded
- The remedy. The credit percentage and its cap. Credits rarely reflect the actual impact on your business, so assess that separately
- The claims process. Whether credits are applied automatically or must be requested within a deadline. Some SLAs give nothing unless you file a claim
- Who measures. Whether uptime is based on the provider’s monitoring alone or your own records can be used too
Setting an SLA for your own service
If you provide a SaaS product and need to set an SLA, work through it in this order:
- List the SLAs you depend on. Write down the SLAs of the cloud, database and external APIs whose failure would take you down. For services chained in series, multiplying their uptimes gives a rough ceiling (two services at 99.9% give about 99.8%)
- Set an internal objective (SLO). Keep an internal target stricter than the SLA, with monitoring that warns you when you are about to miss it
- Decide how to measure. Use external uptime monitoring that checks your service at regular intervals, so you record availability as users see it
- Set the number and exclusions. Decide how scheduled maintenance is announced (how far in advance and where) and which conditions are excluded from uptime
- Define the remedy and claims process. Set the credit percentage, the claim deadline and where claims go
- Publish it. Prepare a public status page and a way to report incidents
How to start monitoring is covered in “What is uptime monitoring?”
How SHANNON can help
As a Google Cloud sales partner, we handle contracts and billing, and design, build and run architectures with the SLA in mind. See our Google Cloud page for details.
References
- Google Cloud: Compute Engine Service Level Agreement (SLA)
- Google Workspace: Service Level Agreement
FAQ
How much downtime does 99.9% uptime allow per month?
In a 30-day month, 99.9% allows about 43 minutes of downtime, 99.95% about 22 minutes and 99.99% about 4 minutes. What counts as downtime, and whether scheduled maintenance is included, differs between SLAs, so read the definitions alongside the number.
If the provider misses its SLA, will it compensate us for our losses?
Most cloud SLAs do not cover losses. Instead they give service credits that reduce future bills by a percentage. Google Cloud's Compute Engine SLA, for example, states that credits are the customer's sole and exclusive remedy. Protect your business through your own architecture and insurance.
What is the difference between an SLA and an SLO?
An SLO (service level objective) is the level a provider aims for. An SLA turns that level into a contractual commitment to customers and defines what happens when it is missed. Teams usually set an internal SLO stricter than the SLA so they notice problems before breaching it.