Special Delivery: Complex Content Delivery for Websites

When customers visit one of Terra Lycos’s websites, they don’t get every object on the pages straight from the online service company’s global Web portal. Instead, a content delivery network (CDN) provider?Akamai Technologies?serves up the images and streaming media from caches (or edge servers, as Akamai calls them) dispersed around the globe. Some localized news even arrives automatically based on which server is nearest to the customer. The result is a lighter load on Terra Lycos’s servers and faster performance for users. But changes to how companies create their pages are forcing CDNs to reevaluate their services.

The original idea behind Akamai and other CDNs was pretty straightforward: Deploy large, often-requested files at the so-called edge of the Internet, where customers could retrieve them quickly. “Back in the days of the bubble, customer experience was the big priority,” says Vince Russo, chief architect at Terra Lycos in Waltham, Mass. “But it also helps us reduce the load to our data centers and the number of servers needed.” In fact, when Lycos started caching its images with Akamai about three years ago, Russo cut his collection of image servers by more than half.

Studies by Giga Information Group have confirmed that CDNs can offer enterprises a significant return on investment, and a reduction in infrastructure costs sounds pretty good during a recession. Unfortunately for CDNs?and their customers?filling repeat requests is getting trickier. “Websites have gotten so much more complicated over the last years,” says Joel Yaffe, a New York City-based analyst with Giga. “It’s not just static content. Now more and more it’s dynamic content, with Web servers asking for complicated information.”

Today, Web servers often build pages on the fly, using content tailored to a customer’s location, type and other ever-changing criteria. It’s a time-consuming process, and one that many companies would like to avoid.

“Companies consider repeat requests an annoying burden on the infrastructure,” says Neal Goldman, an analyst with The Yankee Group in Boston. “If somebody keeps asking you the same question over and over again, someday you’re going to post an FAQ and say, ’Go there, I wrote it down.’ Then you’ll have more resources available to handle nonstandard requests. The question is, How much of that process can you cache?”

The World of CDNs

Both ISPs and companies serving up websites have long used caching to speed up content delivery. Usually, popular content such as homepages and “heavy” files such as graphics and streaming media are placed in a cache for quick access closer to the customer. And customers find the services valuable: The three largest cache providers?Cacheflow, Inktomi and Network Appliance?have combined revenues of approximately $350 million.

What CDNs do is create a distributed set of those caches, with a routing layer that determines which cache is closest to any given website visitor. Some telecommunications companies, including AT&T and Qwest, quietly offer CDN services, but most of the attention goes to the pure plays that are aggressively marketing the CDN model. The largest of them, Akamai, has more than 13,000 servers around the world. The second largest, Digital Island?which purchased rival Sandpiper Networks, and then was purchased by Cable & Wireless?has 2,700 servers. Speedera and Mirror Image are two smaller competitors that remain in the consolidating marketplace.

A new kind of rival is on the way, though. Two different plans in the works would allow companies to link their own CDNs together. One of them is an Inktomi-led initiative called Content Bridge, which would band together smaller CDNs?ideally, ones that used Inktomi boxes?that had fewer nodes than Akamai. The initiative has yet to get off the ground, mainly because many of the CDNs involved went bust along with the Internet bubble. Analysts are more optimistic about the Content Alliance, a self-governing, standards-based group initiated by Cisco Systems.

“What Akamai gives the enterprise is ubiquitous closeness to the consumer,” Goldman says. “If you’re a hosting company, you have your own network, and you look at Akamai and say, ’I can do that, all I need is caching. Why give Akamai any margin?’ But you don’t have the same thing because your network doesn’t reach everywhere in the world.” In response, the Content Alliance is working on an open standard that will allow competing companies to agree to send traffic across each other’s networks, a process called content peering. But creating an open standard is a slow process. Although Content Alliance has an impressive roster?companies such as AOL Time Warner, AT&T, IBM, Qwest, RealNetworks and Sun Microsystems are among the dozens who have signed on?it is still a ways from fruition (too far off to worry about, say some analysts).

Getting Dynamic

In the meantime, existing CDNs see their growth potential in serving up increasingly dynamic content?not just graphics, but information that changes regularly, such as news items.

If Russo has his way, by the time you read this, Terra Lycos will have a whole new model for serving pages. Take, for instance, the news page at news.lycos.com. Different parts of the page have different life spans. The navigation bar rarely changes. But the editor’s picks change every day, and breaking news changes every few minutes. “Each of those regions of the page has a different lifetime,” Russo says. “The fact that part of the page has to be updated every few minutes means that the rest of the page can’t be cached.”

By rewriting the page to a new scripting language called Edge Side Includes (ESI), Russo hopes to change that. ESI is a nonproprietary language developed jointly by Akamai and Oracle. It integrates the applications server with the Akamai network, allowing static webpages to be broken up into different pieces with different lifetimes. The CDN can build the page, pulling information from the originating servers only when necessary. How successful ESI is will depend, ultimately, on how many vendors sign on and the ease with which companies can recode their HTML.

Akamai predicts that by using ESI, Russo will be able to turn off 40 percent to 60 percent of his Web servers. “I’m a little dubious of that number,” he says. “Maybe we’ll only have to serve 6 percent of the content on a page view. But that 6 percent takes a lot of extra power to serve because it’s the most complex part of the page.” Instead, he’s hoping to reduce the number of Web servers by 20 percent or 30 percent, with a smaller reduction in the amount of bandwidth and number of application servers needed.

That may sound good, but CDNs must improve further before they’re dynamic enough for everyone. Kim Ross, CIO of New York City-based Nielsen Media Research, says his company, which crunches data about television audiences, produces data so complex that he hasn’t even Web-enabled it yet?to say nothing of trying to use a CDN to deliver it.

Nielsen’s data changes every day, and few customers want to pull the same information. “Just about every query a customer does is unique to that company?their show against other shows, time period, demographics,” he says. “[Vendors] haven’t made their way to highly dynamic, highly unique, and that’s where the business is.”