Breaking the myth of the nines

Royal Pingdom has posted statistics on the uptimes for the most popular websites in the US for the first quarter of 2007. Apart from Yahoo! who has been up and running all the time, none of the sites on the list made the magic "five nines" mark. The "five nines" mark is the 99.999 % uptime, which translates to 1 minute and 17.5 seconds allowed downtime in a three month period.
Luckily I have seldom seen the "five nine" requirements, but I have often seen requirements for 99.9 % uptime for a wide variety of solutions. Some of these, like some banking systems, could defend this requirements, but most of these apps where non-critical, departmental applications. 99.9 % is about eight and a half hours of downtime per year - or - a little more than two hours per quarter.
Five of the most popular sites in the US would not have met this requirements in the first quarter.

Just to put things in perspective the following list shows how much downtime common percentage requirements allow per year.

• 90% allows 876 hours (36.5 days) of downtime
• 95% allows 438 hours (18.25 days) of downtime
• 99% allows 87.6 hours (3.65 days) of downtime
• 99.9% allows 8.76 hours of downtime
• 99.99% allows 52.56 minutes of downtime
• 99.999% allows 5.256 minutes of downtime
• 99.9999% allows 31.536 seconds of downtime

Few solutions are critical enough to support steep uptime requirements and you should not commit to a requirement you cannot meet. There are a number of reasons why you should lower your expectations on uptime; The money can be spent better on features that add real business value (unless uptime really adds business value) -and- Customers don't really care - Both YouTube and Blogger will remain popular, even if they were down for more than four hours in Q1.

If Google, Microsoft and Amazon either don't bother or don't manage to pull it off, why should you think you will (or will have to)?