99.999% uptime: how much downtime is that?
99.999% uptime, or five nines, allows 0.86 seconds of downtime a day, 26 seconds in an average month and 5 minutes 15 seconds a year. The term comes from telephone networks, which were built to that standard.
Downtime allowed by 99.999% uptime five nines
| Period | Allowed downtime |
|---|---|
| Daily | 0.86s |
| Weekly | 6s |
| Monthly | 26.3s |
| Quarterly | 1m 19s |
| Yearly | 5m 15s |
| 30-day month | 25.9s |
| 31-day month | 26.8s |
| 28-day month | 24.2s |
A month is the average month, a twelfth of a 365-day year; a quarter is a quarter of it.
What that looks like
Restarting one process that every request goes through can take longer than a month's budget. There is no time to notice, decide or act.
Few web services commit to five nines for a whole service in a contract. It is more often a design goal for one component, such as a storage layer or a telephone switch.
How many outages fit
| One outage of | Per month | Per year |
|---|---|---|
| 1m | 0 | 5 |
| 5m | 0 | 1 |
| 15m | 0 | 0 |
| 30m | 0 | 0 |
| 1h | 0 | 0 |
| 4h | 0 | 0 |
Counts whole outages of that length that fit in the budget of an average month or a year, with nothing else going wrong.
What 99.999% takes
- Redundancy at every layer, including DNS, network providers and power.
- Active-active regions with traffic shifted automatically in seconds.
- Changes rolled out one region or one cell at a time, so a mistake reaches only part of the traffic.
- A budget measured over a year; per month there is almost nothing to spend.
Where it fits: Telephone and emergency networks, payment networks and the storage or control layers that large platforms are built on.
How to measure it
A tenth of the monthly budget is 2.6s, shorter than the shortest interval OpenUptime checks at (30 seconds), so checks alone cannot measure a monthly 99.999% precisely. Over a year a tenth of the budget is 31.5s: check at least that often, which in OpenUptime is Team (every 30 seconds), and judge the target over that period.
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 99.999% |
|---|---|---|---|
| 99% | 7h 18m | 3d 15h 36m | 1000× as much |
| 99.5% | 3h 39m | 1d 19h 48m | 500× as much |
| 99.9% | 43m 48s | 8h 45m 36s | 100× as much |
| 99.95% | 21m 54s | 4h 22m 48s | 50× as much |
| 99.99% | 4m 23s | 52m 34s | 10× as much |
| 99.999% | 26.3s | 5m 15s | this page |
Track 99.999% uptime
OpenUptime records every check, shows uptime per monitor and on 90-day status page bars, and alerts you when a check fails.
Questions
How much downtime is five nines?
About 5 minutes 15 seconds a year, 26 seconds in an average month and under a second a day.
Can uptime monitoring prove 99.999%?
Only over a year. A tenth of the yearly budget is about 31 seconds, close to the 30-second minimum interval; over a month a single failed check is already more than the budget.
Is 99.999% worth it?
Each extra nine costs much more than the last in redundancy, tooling and process. It is worth it when an hour of downtime costs more than running the service twice over; for most products 99.9% to 99.99% is the better trade.
99.999% 可用率意味着多少停机时间？
99.999% 可用率，也就是 5 个 9，每天允许 0.86 秒停机，平均每月 26 秒，每年 5 分 15 秒。这个说法来自电话网络，它们就是按这个标准建设的。
99.999% 可用率允许的停机时间 5 个 9
| 周期 | 允许的停机时间 |
|---|---|
| 每天 | 0.86 秒 |
| 每周 | 6 秒 |
| 每月 | 26.3 秒 |
| 每季度 | 1 分 19 秒 |
| 每年 | 5 分 15 秒 |
| 30 天 | 25.9 秒 |
| 31 天 | 26.8 秒 |
| 28 天 | 24.2 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
重启一个所有请求都要经过的进程，花的时间就可能超过一个月的预算。没有时间去发现、判断和处理。
很少有 Web 服务在合同中对整个服务承诺 5 个 9。它更常是某个组件的设计目标，比如存储层或电话交换机。
预算能容纳几次故障
| 单次故障时长 | 每月 | 每年 |
|---|---|---|
| 1 分 | 0 | 5 |
| 5 分 | 0 | 1 |
| 15 分 | 0 | 0 |
| 30 分 | 0 | 0 |
| 1 小时 | 0 | 0 |
| 4 小时 | 0 | 0 |
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 99.999% 需要什么
- 每一层都有冗余，包括 DNS、网络运营商和电力。
- 多区域双活，几秒内自动切换流量。
- 变更按区域或单元逐个发布，出错时只影响部分流量。
- 按年衡量预算；按月几乎没有可以用的余量。
适用场景： 电话和紧急服务网络、支付网络，以及大型平台底层的存储或控制层。
怎样测量
月度预算的十分之一只有 2.6 秒，比 OpenUptime 最短的检查间隔（30 秒）还短，所以仅靠检查无法精确测量按月计算的 99.999%。按年计算时，预算的十分之一是 31.5 秒：检查间隔不要超过这个时长（在 OpenUptime 中对应 Team（每 30 秒）），并按年评估目标。
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 99.999% 相比 |
|---|---|---|---|
| 99% | 7 小时 18 分 | 3 天 15 小时 36 分 | 1000 倍 |
| 99.5% | 3 小时 39 分 | 1 天 19 小时 48 分 | 500 倍 |
| 99.9% | 43 分 48 秒 | 8 小时 45 分 36 秒 | 100 倍 |
| 99.95% | 21 分 54 秒 | 4 小时 22 分 48 秒 | 50 倍 |
| 99.99% | 4 分 23 秒 | 52 分 34 秒 | 10 倍 |
| 99.999% | 26.3 秒 | 5 分 15 秒 | 本页 |
常见问题
5 个 9 允许多少停机？
每年约 5 分 15 秒，平均每月 26 秒，每天不到一秒。
可用性监控能证明 99.999% 吗？
只能按年来衡量。年度预算的十分之一约为 31 秒，接近最短的 30 秒检查间隔；按月来看，一次检查失败就已经超出预算。
99.999% 值得吗？
每多一个 9，在冗余、工具和流程上的投入都比上一个多得多。只有当一小时停机的损失超过把服务多运行一套的成本时才值得；对多数产品来说，99.9% 到 99.99% 是更合适的取舍。