Is the Internet at risk? Not according to Fritz Mueller, senior director of product marketing of Keynote, an Internet performance monitoring firm. Mueller uses this analogy: “If you take out an automated teller machine [ATM], does this mean that the entire banking system is gone? No, it just means that the people who want to access the content or services provided by that ATM cant.”
So, if an exchange or peering point, or a co-location facility, is damaged or otherwise “taken out,” the chances of a complete breakdown are marginal. But Muellers analogy has another facet. If that ATM is the only place you can get money, then youre in a world of trouble. Mueller says hes more concerned about replicating content in multiple locations than he is about the Internet failing.
Adds Keynote VP of operations Lloyd Taylor, “Im more worried about the potential lack of power [in California] this spring than I am about the Internet failing.” While Taylor believes that most facilities have sufficient generators to guarantee several days of self-sufficiency, he wonders where all of the fuel will come from. “The gas pipelines stop pumping when theres no power,” he says.
If a large natural disaster were to occur in the Bay Area, Taylor says the network would re-route traffic. Still, he adds, “If your data were only in one location in San Jose, then you have some issues.”
And therein lies the crux of the problem. Integratus, a high-availability software provider that provides site fail-over across the Internet, advises customers to look for single points of failure throughout their infrastructures. “Youve got to worry about server availability, network availability and site availability,” says CEO Phil Kufeldt. “In a typical e-commerce application, youll want to ensure redundancy at the access tier, the content creation and the data itself.”
While many vendors—including Sun, Oracle, Veritas, Akamai, and EMC—provide solutions to the problem, one vendor is trying to provide a full software infrastructure for distributed Internet applications. Zembu Labs, a startup in Palo Alto, Calif., recently unveiled its distributed application infrastructure, which distributes existing applications, including the database, to multiple geographic locations. Users then are directed to the instance of the application that provides the quickest response. Zembus algorithms ensure that data is properly propagated and synchronized to each database without sacrificing integrity or security.
Its unfortunate that theres no such thing as high-availability insurance for e-commerce systems. So what do you do? You can take the position of Josh Cates, CTO of content delivery network provider Scale8: “Everything breaks, all the time,” he says.
With that in mind, build your site(s) accordingly.