SLO burn-rate example: comparing one hour and five minutes

Is the higher error rate visible in both windows?

Both illustrative windows have 1.5% bad events against a 0.1% allowed fraction, producing 15× burn. The long window has 1,500 bad events out of 100,000 over one hour; the short window has 150 out of 10,000 over five minutes. Both exceed the example 14.4× checkpoint. Over a 30-day period, the long window represents about 2.08% normalized budget spend.

What this example does not establish

Windows must end together and use the same SLI. This demonstrates threshold arithmetic, not a deployed alert or measured remaining rolling budget. Variable traffic changes the interpretation of normalized spend.

How a different input changes the result

Reduce short-window bad events from 150 to 100 and leave its 10,000 eligible events unchanged. The short rate becomes 10× while the long rate remains 15×. Only the long window now exceeds the 14.4× checkpoint, which explains why the two-window condition no longer holds.

Read the related guide · Source and further reading

These datasets and worksheets may be adapted for your team or training. Attribute AIOpsSRE and link to this example when redistributing. Third-party sources retain their own terms.

Get AIOpsSRE by email · Report a calculation or explanation issue