Measured #1 Most Reliable


May 13, 2014 by Brian Ghidinelli

In the hosting industry, the slang for reliability is the number of "nines" you run. Most web hosting companies claim something like "five nines", which equates to 3 minutes and 48 seconds of downtime per year. Companies like Rackspace will sell you a zero-downtime network which sounds great but your only recourse for the eventual outage (oh, and there will be one or two) is a few pennies of your hosting dollars. Most serious services aim, realistically, for four nines or 99.99% uptime.

At Pukka Software, we take the reliability of seriously. The problem is every company says they care about reliability but few share their numbers. Until today! Using data provided by impartial, third-party monitoring services, these charts compare the last 90 days between us and three services whom our customers previously used before switching to

For the purpose of this discussion, we care only about service availability. That is, if I type in the address into my web browser, does the website answer? It's quite possible for the system to be "up" but the network or some other resource to be "down". We care only if the website responds.

This first chart shows uptime expressed as a percentage. Most people look at a chart like this and say: "Woopity doo! Even the least reliable service,, is running 98.9215% of the time! That's pretty good!"

That is true. Maybe. How frequent are the outages? How long is the duration? Imagine a website that is regularly down for short periods of time. These flaky websites are typically caught out by unexpected, non-catastrophic issues. Now consider a website that suffers a 6, 12 or 24-hour outage. These kinds of massive outages are almost always the result of poor strategy and contingency planning. In the end, neither type is particularly appealing as a user.

So what is the practical difference between our 99.9288% and that of our competitors? This second chart switches from percentage to absolute number of minutes offline in the past 90 days. and have more than 1,300 minutes each of downtime in the past 90 days. Even, who has made improvements in their reliability in past months, was unavailable three times as often compared to

"To provide 99.9% or better uptime, you'll need to house your equipment in an environmentally controlled space with guaranteed 100% power availability. In most cases this will be accomplished by purchasing colocation space in a top-tier data center... You will need redundant equipment for all of your key systems... Three nines of reliability is very achievable, but it takes considerable expense and requires a fair degree of competency in your staff."

The rule of thumb with "nines" of availability is that each additional nine costs an order of magnitude more money. In my experience this is true. We know where these operations host their websites and there is a direct correlation between the amount invested and the results in these charts. With our internal target of "four nines" at 99.99%, or 60 minutes of downtime per year, these past three months were regrettably a ding against our record as our historical reliability is a very respectable 99.963%. This includes our physical colocation move last year which resulted in a planned, early-morning outage to migrate our hardware into a private, locking cabinet for better customer data security and compliance with PCI DSS.

At 99.963%, is unavailable for just three hours per year. No other competing service provider has a historical uptime even in the "triple nines" at 99.9%. It's not as good as the four nines we're striving for but we are building on a solid track record in its pursuit.

Subscribe to Blog Updates

Subscribe by RSS

Follow MSR

Like us on Facebook Connect with us on LinkedIn Follow us on Twitter

Recent Posts