What’s With All The Nines? (Availability)

If you are a user of cloud platforms (e.g. AWS, Azure, Google Cloud) you have probably noticed the mentions of availability that include a lot of nines (9s). Generally these are 99.xx (ninety-nine point something), and the more nines after the ninety-nine point means more availability but to what level?

If the service was available 100% of the time that would be 24hrs a day, 7 days a week, 52 weeks a year (e.g. 365 days a year, unless it is leap year). So 365 days equals 100% uptime on a yearly schedule, i.e. no unavailability. However, what does five nines (99.999%) mean in terms of unavailability?

Percentage of AvailabilityUnavailability
90%36.5 days
95%18.25 days
99%3.65 days
99.9%8.76 hours
99.99%52.6 minutes
99.999%5.26 minutes
Yearly Availability

99.999% availability would be just under 5 and half minutes of unavailability every year, quite a feat of engineering. You may be wondering what the five nines (99.999%) means for unavailability on a monthly term. If we take a month as 30 days (yes some have 31 and one has 28 or 29 depending on the year but let’s round it to 30) then:

Percentage of AvailabilityUnavailability
90%3 days
95%1.5 days
99%7.2 hours
99.9%43.2 minutes
99.99%4.32 minutes
99.999%25.9 seconds
Monthly Availability

99.999% of availability is under 30 seconds of availability a month. I’ve worked with computers that have taken longer than 30 seconds to boot (although big cloud platforms have lots of redundancy built into their systems and won’t be relying on just 1 device).

On an hourly basis the availability / unavailability figures scale down to “blink and you’ll miss it” levels.

Percentage of AvailabilityUnavailability
90%6 minutes
95%3 minutes
99%36 seconds
99.9%3.6 seconds
99.99%0.36 seconds
99.999%0.04 seconds
Hourly Availability

99.999% hourly availability is 0.04 seconds (slight rounding) of unavailability an hour. That’s point zero four of a second. Would that even be noticeable to most users when factors away from the cloud provider such as the end users network/ISP, end users computer etc… are factored in?

The above figures are all on the 24/7/365 scale. If the organisation / platform offers 99.999% on a different scale (e.g. core office hours of 9am to 5pm) then the unavailability times would alter.