99.9% uptime: how much downtime is that?
99.9% uptime, or three nines, allows 1 minute 26 seconds of downtime a day, 43 minutes 48 seconds in an average month and 8 hours 45 minutes a year. It is one of the most common figures in SaaS and cloud SLAs.
Downtime allowed by 99.9% uptime three nines
| Period | Allowed downtime |
|---|---|
| Daily | 1m 26s |
| Weekly | 10m 5s |
| Monthly | 43m 48s |
| Quarterly | 2h 11m 24s |
| Yearly | 8h 45m 36s |
| 30-day month | 43m 12s |
| 31-day month | 44m 38s |
| 28-day month | 40m 19s |
A month is the average month, a twelfth of a 365-day year; a quarter is a quarter of it.
What that looks like
One 45-minute incident uses the whole month. Two deploys that each break the site for 20 minutes come close to it.
Whether the target is monthly or yearly matters: a yearly 99.9% can absorb one 8-hour outage, a monthly one cannot.
How many outages fit
| One outage of | Per month | Per year |
|---|---|---|
| 1m | 43 | 525 |
| 5m | 8 | 105 |
| 15m | 2 | 35 |
| 30m | 1 | 17 |
| 1h | 0 | 8 |
| 4h | 0 | 2 |
Counts whole outages of that length that fit in the budget of an average month or a year, with nothing else going wrong.
What 99.9% takes
- On-call paging, with someone looking at an alert within about 10 to 15 minutes.
- Deploys that can be rolled back in a few minutes.
- No single machine whose failure takes the service down for long: a database replica you can fail over to, more than one app instance.
- Checks every minute, from more than one region, so a real outage is caught quickly and a single network blip does not page anyone.
Where it fits: Customer-facing SaaS, public APIs and online shops: the usual promise to paying customers.
How to measure it
Check at most every tenth of the budget, so short outages do not slip between checks: every 4m 23s or more often for a monthly 99.9%. In OpenUptime that is Lite or Starter (every minute) or Team.
Check from more than one region, so a single network problem between one location and your server is not counted as downtime.
Compared with other targets
| Uptime | Monthly | Yearly | Downtime against 99.9% |
|---|---|---|---|
| 99% | 7h 18m | 3d 15h 36m | 10× as much |
| 99.5% | 3h 39m | 1d 19h 48m | 5× as much |
| 99.9% | 43m 48s | 8h 45m 36s | this page |
| 99.95% | 21m 54s | 4h 22m 48s | 1/2 as much |
| 99.99% | 4m 23s | 52m 34s | 1/10 as much |
| 99.999% | 26.3s | 5m 15s | 1/100 as much |
Track 99.9% uptime
OpenUptime records every check, shows uptime per monitor and on 90-day status page bars, and alerts you when a check fails.
Questions
Is 99.9% uptime per month or per year?
Whatever the SLA says. Most cloud and SaaS SLAs measure per month, which is stricter: 43 minutes a month cannot be saved up for one long outage.
How many nines is 99.9%?
Three nines. 99.99% is four nines and 99.999% five.
What happens when a 99.9% SLA is missed?
Usually the customer can claim a service credit, a share of that month's fee. The size of the credit and how to claim it vary by contract.
99.9% 可用率意味着多少停机时间？
99.9% 可用率，也就是 3 个 9，每天允许 1 分 26 秒停机，平均每月 43 分 48 秒，每年 8 小时 45 分。这是 SaaS 和云服务 SLA 里最常见的数字之一。
99.9% 可用率允许的停机时间 3 个 9
| 周期 | 允许的停机时间 |
|---|---|
| 每天 | 1 分 26 秒 |
| 每周 | 10 分 5 秒 |
| 每月 | 43 分 48 秒 |
| 每季度 | 2 小时 11 分 24 秒 |
| 每年 | 8 小时 45 分 36 秒 |
| 30 天 | 43 分 12 秒 |
| 31 天 | 44 分 38 秒 |
| 28 天 | 40 分 19 秒 |
一个月指平均月，即 365 天的十二分之一；一个季度是一年的四分之一。
实际意味着什么
一次 45 分钟的故障就会用完整个月的预算。两次发布各让网站中断 20 分钟，也就差不多用完了。
目标按月还是按年计算差别很大：按年计算的 99.9% 能容纳一次 8 小时的故障，按月计算的就不行。
预算能容纳几次故障
| 单次故障时长 | 每月 | 每年 |
|---|---|---|
| 1 分 | 43 | 525 |
| 5 分 | 8 | 105 |
| 15 分 | 2 | 35 |
| 30 分 | 1 | 17 |
| 1 小时 | 0 | 8 |
| 4 小时 | 0 | 2 |
表示在平均月或一年的预算内，能容纳多少次这样时长的完整故障（假设没有其他问题）。
做到 99.9% 需要什么
- 有值班告警，告警发出后大约 10 到 15 分钟内有人处理。
- 发布出问题时能在几分钟内回滚。
- 没有哪台机器一出故障就让服务长时间中断：数据库有可以切换的副本，应用不止一个实例。
- 每分钟检查一次，并从多个区域检查，这样真正的故障能很快发现，单次网络抖动也不会把人叫醒。
适用场景： 面向客户的 SaaS、公开 API 和网店：对付费客户的常见承诺。
怎样测量
检查间隔不要超过预算的十分之一，以免短暂的故障落在两次检查之间：按月计算的 99.9%，检查间隔应不超过 4 分 23 秒。在 OpenUptime 中，这对应 Lite 或 Starter（每分钟）或 Team。
从多个区域检查，这样某一个地点到服务器之间的网络问题就不会被算作停机。
与其他目标对比
| 可用率 | 每月 | 每年 | 停机时间与 99.9% 相比 |
|---|---|---|---|
| 99% | 7 小时 18 分 | 3 天 15 小时 36 分 | 10 倍 |
| 99.5% | 3 小时 39 分 | 1 天 19 小时 48 分 | 5 倍 |
| 99.9% | 43 分 48 秒 | 8 小时 45 分 36 秒 | 本页 |
| 99.95% | 21 分 54 秒 | 4 小时 22 分 48 秒 | 1/2 |
| 99.99% | 4 分 23 秒 | 52 分 34 秒 | 1/10 |
| 99.999% | 26.3 秒 | 5 分 15 秒 | 1/100 |
常见问题
99.9% 的可用率是按月还是按年算？
以 SLA 的约定为准。多数云服务和 SaaS 的 SLA 按月计算，这样更严格：每月 43 分钟的预算不能攒起来用在一次长时间故障上。
99.9% 是几个 9？
3 个 9。99.99% 是 4 个 9，99.999% 是 5 个 9。
达不到 99.9% 的 SLA 会怎样？
通常客户可以申请服务抵扣，即当月费用的一部分。抵扣多少、怎么申请，各家合同不同。