Service level agreements: what a monthly average hides
A monthly availability number can hide a bad minute. Google's Site Reliability Engineering book defines availability as the fraction of time a service is usable, often the fraction of well-formed requests that succeed.
It says 100 percent availability is impossible, and it uses 99 percent as the example of "2 nines." It also says raw measurements are aggregated over a window into a rate, average, or percentile, and that an average can hide a burst: a system that serves 200 requests a second in one second and none in the next has the same average as a steady 100, with twice the instantaneous load.
An error budget is the rate at which a service level objective is allowed to be missed.
An hour on a busy morning may not move a monthly figure that is still above a line such as 95 percent. A short service credit can also be small next to the time the business was without the service. Those are reasons to read an average carefully. They are not a current measurement of this site and not a puna.one price.
Does this page pay a credit when a site is down?
No. This page does not sell an uptime credit. Google's book says a public service can have no SLA with the world, using Google Search as its example, while other services do have explicit SLAs with their users. The work people actually need is someone watching the site and backups that can be restored.
Current hosting price and scope are on the managed hosting page. This guide does not add an uptime percentage.
Is 99 percent the current uptime?
No. Google's book treats 99 percent as an illustration of how availability targets are named, not as this site's target.
Related reading
Key takeaways
- A monthly availability average can hide a short outage.
- This page does not sell an uptime credit and does not add an uptime percentage.
- 99 percent is an illustration of how availability targets are named, not this site's target.
Next step: Book a Managed Services intro.