In talking about server uptime the discussion has come up as to what it means to say “5-9s” or “6-9s”.  Each has a VERY specific meaning in regards to the average percentage uptime in any given 365 day period.  5-9s means a 99.999% uptime (there’s 5 9s).  That’s 364.99635 days, or better put only 5 minutes and roughly 15 seconds of downtime in the aggregate over 365 days.

If you’ve ever had a webserver that got hit by the slashdot/digg/fark/reddit effect, then you’d know that your host probably hates you for destroying their 5-9 uptime.  On one of my other sites, I had an article that was frontpaged on reddit for a few hours and linked to a good number of popular news sites.  In 48 hours, I recorded over 150,000 unique visits, and all three of my tracking mechanisms were down for an extensive period.  The traffic brought down the entire server for about 2 hours, and that was after the page was changed to static HTML and I dumped a ton of the markup on it to remove linked js libraries and images.  My host moved me to a much less used and tragically faster server as a result, and the hosting packages is now one of the best around for non-colo.  In the end, I learned a lot about servers under high load.

Now, here’s a table to help understand exactly how much downtime is allowed before the threshold is breached.

Note that the units on the far right column change.  For a day unit, the max downtime decimal is base 24 in hours.  So for instance, 75% uptime is 91.25 days, but .25 days is 6 hours, so it’s 91 days and 6 hours.  At 3-9s and below, the decimal is base 60, so the max downtime is 8 hours 45 minutes and 36 seconds for 3-9s.

If you’re really a stickler, you have to pay attention to how the service provider phrases the uptime claim.  If they say “we provide better than 99.999% uptime” (as opposed to “we provide 99.999% uptime or better”), that means there’s LESS than but not equal to 5.256 minutes of downtime in a 365 day period.  It’s really just an inclusive/exclusive question on the range of  numbers.

As for choosing the right level of service for your site, it really comes down to number crunching and how much your site actually brings in.  Take, for example, the blog BoingBoing.  Federated Media sells out a 125×125 ad spot on the site for $7.00 CPM.  With 7.83 million monthly page views, that means the site gets a steady 3 hits per second.  Even if we say that the site only has a single 125px square for ad revenue on each page to generate revenue (though it has more), the site brings in AT ABSOLUTE VERY LEAST 2.1 cents per second on average.  The difference between 6-9s and 7-9s is about 28 seconds, so you can see that at only 2.1 cents per second, that extra 9 only brings in another 60 cents.  Now because we can only compare the bottom end of the income (as opposed to the top end of the income) to the service costs we can only say what might be a good deal - not what would be a bad deal (with more data on actual income per second, we could determine to the second how much guaranteed uptime they should buy).  The difference between 6-9s and 7-9s is only about 28 seconds.  So if upgrading from 6-9s to 7-9s costs 60 cents or less (incredibly unrealistic), we can safely say that’s a good deal.  But we’re not sure 6-9s is even worth it, so lets go a little farther down the uptime list to 1-9 vs 2-9s.  At 1-9, that’s 36.5 days of downtime.  In 36.5 days downtime at 2.1 cents per second, the site misses out on at least $66,225.60.  At 2-9s, the site misses out on $6,622.56.  Pragmatically, if the cost between going from 1-9 to 2-9s is less than the difference ($59,604.04), then it’s definitely worth it for the company to upgrade service from 1-9 to 2-9s.

For companies that use their site solely for referrals that convert into a sale or service contract, 4-9 or 5-9 service levels (which are rather common) are typically more than enough.  But, in very high end enterprise systems like investment companies, stock exchanges, and multi-billion record inventory databases where the income per second is measured in the hundreds or thousands or even millions of dollars, even 15-9s uptime might be worth the fees.