It's tough to build a strong Web foundation on these shifting sands. You can choose the best hosting provider on the planet and still not be able to reach a good part of it consistently. No matter how good the data center services, no matter how wide the backbone, some clients somewhere will want your content but won't be able to get it. Clearly, we need a different way of serving up Web pages that accepts the reality of bad traffic management in the Internet.
Content delivery networks (CDNs) are a proven solution to the problem. Pioneered by Akamai Technologies' FreeFlow, CDNs reduce the network load and the strain on back-end servers by pushing content to servers that are "close" to the remote client. And the approach works: Some of the most highly trafficked Web sites--including those for Yahoo!, Lycos, Go Network and The New York Times--use CDNs. These sites support enormous loads, yet they consistently end up as the fastest sites in Keynote Systems' performance indices (www.keynote.com).
CDNs do a great job pushing out content from publishing-oriented Web sites, as well as offering some support for streaming media. What they don't do particularly well is push transactions to the edge of the network, often requiring the centralized Web server to connect to a back-end application engine before closing the deal. And though they reduce failures caused by overloaded servers and networks, CDNs don't replace localized server load-balancing techniques. If the back-end server dies, you're still out of business.
A CDN must adjust rapidly to changes in traffic patterns on the Internet. The perfect server for delivering content may change at a moment's notice--because of packet loss, delay, route changes or the load on the cache server. Some ISP-based CDN services attempt to count the number of hops across the network, but that's not intelligent enough to consistently deliver good service.
To react to tectonic shifts in Internet traffic patterns, CDN providers must build facilities cheaply. This is especially true outside of North America, where traffic characteristics vary greatly by country. Here the pure network-based overlays, such as InterNAP and AboveNet, and the satellite-based players, such as iBeam and Cidera, simply can't move fast enough to keep up with the Internet's changes; the required investment is too high to build quickly. In contrast, the server-based CDNs (Akamai, Digital Island/Sandpiper, Adero, Mirror Image) simply add remote servers. More important, they add these servers to both their own and other service provider networks, so they don't have to rely on a single network infrastructure for delivery. If one content server--or even an entire backbone--is down, clients are redirected to another.
While CDNs can give your site a huge edge in performance, you must be careful how you deploy them. All the major CDNs charge for usage, and you may be hard-pressed to predict your traffic patterns--especially in the early stages of an initiative.
There are many emerging competitors in the content-delivery space. To survive, CDN services must expand beyond cache-based delivery to offer application logic and point of interaction customization.
David Willis is a program director for the Meta Group's Global Networking Strategies service.
Send your comments on this column to him at david.willis@metagroup.com.
The Internet will forever be congested. Always has been, always will be. No matter what counteractive measures we take--fatter fiber pipes, faster routing, better peering or route aggregation--the Internet will still need a big dose of Flonase. According to recent MIDS data (www.mids.org), many major providers aren't reachable 10 percent of the time. Others routinely drop more than a tenth of their traffic. A rare few aspire to the world-class packet-drop records held by Netcom/MindSpring, which in a recent week dropped one quarter of its packets and wasn't even there 13 percent of the time.
REPORTS
ANALYTICS
Follow 
