How long does a start take? Here is what we measure.

How we measure, and what this does not tell you

• What is measured: the time between your start request and the moment your server is announced as “ready”, for starts that succeeded. Starts that fail are not in these figures.
• The median is the time that half of the starts beat. The “9 in 10” column (the p90) shows the bad days: nine starts out of ten are faster than that duration.
• The country is the one you pick for your server. A cell shows a figure only from 10 measurements: below that we write “not enough measurements” rather than a fragile number.
• “New machine”: a machine prepared for your start. “Resume”: a machine already running, reassigned to your server, hence faster.
• Internal accounts and acceptance-test runs are excluded. The figures are recalculated once a day, so they lag by 24 hours at most.
• This is not a commitment about your next start: it is what happened. And this waiting time is not billed, the first hour begins when your server is ready.