How many nines?

Or do you really need high availability?

Every time I have a conversation with a client about bootstrapping a new deployment, they raise the prospect of high availability — failovers, and multiple availability zones, and all that jazz.

AWS likes to sell you all that, so it’s front and centre on most of the pages they will have seen.

Most useful page on the internet of this conversation is Wikipedia’s on High availability, and in particular the Percentage calculation table showing how much downtime you’re allowed for various numbers of “nines” – where 90% is one-nine, 99% two-nines, and so on.

The bottom-line is that on the simplest, no-failover deployment you’ll get likely four-nines, and comfortably three-nines without even blinking.

Most quarters you’ll have zero-downtime, so quite a few nines there.

More, you’ll get that at a fraction of the infrastructure and ops cost of the next level up.

Usually that’s the end of the conversation, but having the table to look at is a real aid. I have it bookmarked for quick access.

(Considerations are different for high-traffic monsters, but those are few and far between in honesty.)