99.95% uptime: how much downtime is that?
99.95% uptime allows 43 seconds of downtime a day, 21 minutes 54 seconds in an average month and 4 hours 22 minutes a year. It sits between three and four nines, with half the budget of 99.9%.
Downtime allowed by 99.95% uptime
| Period | Allowed downtime |
|---|---|
| Daily | 43.2s |
| Weekly | 5m 2s |
| Monthly | 21m 54s |
| Quarterly | 1h 5m 42s |
| Yearly | 4h 22m 48s |
| 30-day month | 21m 36s |
| 31-day month | 22m 19s |
| 28-day month | 20m 10s |
A month is the average month, a twelfth of a 365-day year; a quarter is a quarter of it.
What that looks like
A single 20-minute incident uses almost all of a month. By the time someone is paged, has opened a laptop and found the cause, much of the budget is gone.
Deploys have to be invisible: a release that takes the service down for a minute, every day, would use 30 minutes a month.
How many outages fit
| One outage of | Per month | Per year |
|---|---|---|
| 1m | 21 | 262 |
| 5m | 4 | 52 |
| 15m | 1 | 17 |
| 30m | 0 | 8 |
| 1h | 0 | 4 |
| 4h | 0 | 1 |
Counts whole outages of that length that fit in the budget of an average month or a year, with nothing else going wrong.
What 99.95% takes
- Instances spread over several availability zones, with the load balancer taking failed ones out.
- Rolling or blue-green deploys, so a release never stops the service.
- A database that fails over to a replica on its own.
- On-call that responds within minutes, and checks every 30 seconds to a minute.
Where it fits: Platforms other businesses build on, payment and checkout flows, and services sold with a stronger SLA than the usual 99.9%.
How to measure it
Check at most every tenth of the budget, so short outages do not slip between checks: every 2m 11s or more often for a monthly 99.95%. 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 99.95% |
|---|---|---|---|
| 99% | 7h 18m | 3d 15h 36m | 20× as much |
| 99.5% | 3h 39m | 1d 19h 48m | 10× as much |
| 99.9% | 43m 48s | 8h 45m 36s | 2× as much |
| 99.95% | 21m 54s | 4h 22m 48s | this page |
| 99.99% | 4m 23s | 52m 34s | 1/5 as much |
| 99.999% | 26.3s | 5m 15s | 1/50 as much |
Track 99.95% uptime
OpenUptime records every check, shows uptime per monitor and on 90-day status page bars, and alerts you when a check fails.
Questions
Why pick 99.95% instead of 99.9% or 99.99%?
It is a step up from 99.9% that a team with redundancy and automatic failover can still meet with people in the loop. 99.99% leaves about 4 minutes a month, too little for anyone to respond by hand.
How much downtime is 99.95% per month?
About 21 minutes 54 seconds in an average month, 21 minutes 36 seconds in a 30-day month and 22 minutes 19 seconds in a 31-day month.
Are 1-minute checks enough to measure 99.95%?
Yes. A tenth of the monthly budget is about 2 minutes 11 seconds, so 1-minute checks keep the measuring error small; 30-second checks make it smaller.
99.95% 可用率意味着多少停机时间？
99.95% 可用率每天允许 43 秒停机，平均每月 21 分 54 秒，每年 4 小时 22 分。它介于 3 个 9 和 4 个 9 之间，预算是 99.9% 的一半。
99.95% 可用率允许的停机时间
| 周期 | 允许的停机时间 |
|---|---|
| 每天 | 43.2 秒 |
| 每周 | 5 分 2 秒 |
| 每月 | 21 分 54 秒 |
| 每季度 | 1 小时 5 分 42 秒 |
| 每年 | 4 小时 22 分 48 秒 |
| 30 天 | 21 分 36 秒 |
| 31 天 | 22 分 19 秒 |
| 28 天 | 20 分 10 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 20 分钟的故障几乎就会用完一个月。等值班的人收到告警、打开电脑、找到原因，预算已经用掉了一大半。
发布必须对用户无感：如果每天发布一次、每次中断一分钟，一个月就是 30 分钟。
预算能容纳几次故障
| 单次故障时长 | 每月 | 每年 |
|---|---|---|
| 1 分 | 21 | 262 |
| 5 分 | 4 | 52 |
| 15 分 | 1 | 17 |
| 30 分 | 0 | 8 |
| 1 小时 | 0 | 4 |
| 4 小时 | 0 | 1 |
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 99.95% 需要什么
- 实例分布在多个可用区，负载均衡器会把故障实例移出。
- 滚动发布或蓝绿发布，发布时服务不中断。
- 数据库能自动切换到副本。
- 值班人员几分钟内响应，检查间隔为 30 秒到 1 分钟。
适用场景： 其他企业在其上构建业务的平台、支付和结账流程，以及承诺比常见 99.9% 更高 SLA 的服务。
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 99.95%，检查间隔应不超过 2 分 11 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 99.95% 相比 |
|---|---|---|---|
| 99% | 7 小时 18 分 | 3 天 15 小时 36 分 | 20 倍 |
| 99.5% | 3 小时 39 分 | 1 天 19 小时 48 分 | 10 倍 |
| 99.9% | 43 分 48 秒 | 8 小时 45 分 36 秒 | 2 倍 |
| 99.95% | 21 分 54 秒 | 4 小时 22 分 48 秒 | 本页 |
| 99.99% | 4 分 23 秒 | 52 分 34 秒 | 1/5 |
| 99.999% | 26.3 秒 | 5 分 15 秒 | 1/50 |
常见问题
为什么选 99.95%，而不是 99.9% 或 99.99%？
它比 99.9% 更进一步，而有冗余和自动故障切换的团队，在仍需人工参与的情况下也能做到。99.99% 每月只剩约 4 分钟，来不及人工响应。
99.95% 每月允许多少停机？
平均月约 21 分 54 秒，30 天的月 21 分 36 秒，31 天的月 22 分 19 秒。
1 分钟检查一次，足以测量 99.95% 吗？
足够。月度预算的十分之一约为 2 分 11 秒，所以每分钟检查一次误差已经很小；每 30 秒检查一次会更小。