Skype (eBay), Cisco, salesforce.com, Yahoo, Google, Netflix, Six Apart, Craigslist, ZDNet and other CNET site--the list goes of of sites that have suffered more than a few hours of downtime downtime for various services. The causes vary, from 365 Main's failed backup system to "software problems," as Skype explained its more than 24-hour outage this week.
These kinds of failures to deliver services are representative of the state of the Internet. The Internet is inherently redundant and robust, but you cannot count on most services being available for five nines, which translates to a down time of 5.26 minutes per year. And, that calculation doesn't include times when a site performs slowly and is almost unusable.
We are at the beginning of a 20-year transition toward utility computing, when massively scaled, energy efficient, five-nines compute, storage and networking services power the Internet. Today, we are hamstrung by the complexity and unwieldy nature of the systems in place. In addition, you get what you pay for, and since it's mostly free or very cheap, you're getting three nines of great performance if you are lucky. But it's not hard to imagine that the complexity and problems with today's systems can be overcome with money and more intelligence in the systems, especially for warding off the bad bots.