Availability table: allowed downtime from 90% to 99.999% uptime
See how much downtime 99.9%, 99.99%, and 99.999% availability allow. Calculate time-based error budgets and distinguish them from failed-request budgets.
At 99.9% uptime, a service can be down for 43.2 minutes in 30 days. At 99.99%, the same window allows 4.32 minutes. If a failover takes ten minutes, that one incident exceeds the entire 30-day budget at the higher target, even if everything else works.
Those numbers apply to time-based availability. If your service-level objective (SLO) measures successful requests, count failures against the request budget. Use the availability table to translate a time-based target into downtime, then check that its assumptions match the way you measure the service.
Uptime and allowable downtime table
| Availability | 365 days | 30 days | 7 days | 24 hours |
|---|---|---|---|---|
| 90% | 36.5 days | 3 days | 16.8 hours | 2.4 hours |
| 99% | 3.65 days | 7.2 hours | 1.68 hours | 14.4 minutes |
| 99.5% | 1.83 days | 3.6 hours | 50.4 minutes | 7.2 minutes |
| 99.9% | 8.76 hours | 43.2 minutes | 10.1 minutes | 1.44 minutes |
| 99.95% | 4.38 hours | 21.6 minutes | 5.04 minutes | 43.2 seconds |
| 99.99% | 52.6 minutes | 4.32 minutes | 1.01 minutes | 8.64 seconds |
| 99.999% | 5.26 minutes | 25.9 seconds | 6.05 seconds | 0.86 seconds |
Multiply the window duration by one minus the target expressed as a fraction. For 99.9% over 30 days, that is 43,200 minutes × (1 − 0.999) = 43.2 minutes. These are rounded conversions for fixed windows. A calendar month can have a different length. Use the dates, time zone, and exclusions in your actual measurement rules for reporting. Google SRE availability table
Put the same denominator in the objective and the alert
A time-based service-level indicator (SLI) counts how much of the window met your definition of available. A request-based SLI might divide successful checkout requests by all eligible checkout requests. At 99.9% across 2 million eligible requests, the budget is 2,000 failed requests. Ten busy minutes can consume that allowance faster than ten quiet minutes. The downtime table cannot describe that difference.
Decide whether a successful response must also meet a latency threshold. An endpoint can return HTTP 200 after 30 seconds and still fail a success criterion that includes latency. You can also keep the availability and latency objectives separate. Define which routes, users or regions are included, then check whether your service-level agreement (SLA) uses the same scope and whether it excludes planned maintenance. Google SRE workbook on implementing SLOs
What another nine asks of recovery
A ten-minute outage consumes about 2.3 times the budget for one 30-day window at 99.99%. It does not draw down the budget for the next independent calendar period. On a rolling window it continues to count until it ages out. At 99.999%, even a 365-day window allows just over five minutes, so the time to detect, fail over, and recover matters as much as the redundancy on the architecture diagram.
The additional cost depends on which failures still affect users. If recovery waits for a manual database promotion, start by measuring that path. If both sites share the same dependency, another site may leave the limiting failure untouched. Also keep incident count and duration on the review page. One 43-minute outage and 43 one-minute outages spend the same time budget if all fall in the window, but can disrupt users in different ways.
Measure the experience you promised
An external HTTP probe can reveal a DNS or routing failure while the server's own health check still passes. In Guance synthetic monitoring, choose locations that reflect your users and configure checks for the response you expect. The pass rate is measured over test results, and a short outage can occur between checks. Decide how you treat missing results and regional failures before using probes to support an SLA report.
In Guance SLO, Monitor type uses the duration covered by monitor anomalies. Metric type compares successful counts with total counts. Use the type that matches your SLI and define the window and scope before setting an alert. How quickly the alert fires depends on the monitoring configuration.
An error budget is the amount of failure the objective allows. Spend it in minutes or failed requests according to the SLI. When it runs low, use the team's agreed policy to decide which changes to pause and which reliability work to prioritize. The table tells you how much room you have. Detection and recovery times tell you whether that room is enough.
Contact us
Join the community
to join the community
Try Guance
Start online and pay only for what you use.
Get startedChoose a Guance plan
