Bringing it home

Content delivery has become a new battleground - and maybe the commodity of the future

All the maxims about content on the Internet - that it's king, that it wants to be free, that it drives broadband adoption and so on - bow down before one central truth: that it can't do squat if it's not delivered.

Everyone who's ever clicked has known the irritation of a Web page that takes a minute to load or that never loads at all. Such errors are also a headache, of course, for the Web site and the dot-com that owns it. But re-liable, speedy content delivery is becoming a prime concern for service providers as well, especially those that own their own backbone networks.

The next few years will see a swelling of the bandwidth glut that will drive connectivity prices into the cellar if someone doesn't find a use for all that empty fiber. Renaissance Worldwide estimates that 64% of the lit European net capacity will go unused in the next five years; in the U.S., that proportion will hit 38% - still sizeable. The search for a bandwidth-hungry killer app is on.

"Connectivity is accomplished," says Jonathan Seelig, vice president of strategy and corporate development for Akamai Technologies. "With some exceptions, the infrastructure is falling into place. Now service providers are having to concern themselves with what goes over those lines. Content - the ability to deliver it quickly and well - will become the next access."

Multimedia seems destined to be that app for the broadband age, just as e-mail was for the larval phase of the Internet. But content applications face the hurdle of the last mile. The Internet was built to send e-mail reliably, not real-time footage of Christy Turlington slinking down a catwalk. The vagaries of dial-up access - latency, jitter - only compound the problems.

Yet Web sites keep packing their pages with muscular multimedia applications. A review of 1000 Web sites during second quarter 2000 by Decise, a Web intelligence consultant, revealed that 61% of the largest sites - those with more than 1000 pages - offer audio, video or Flash multimedia content.

The market associated with content delivery will be large - $1.4 billion in 2003, according to The Yankee Group (Figure 1).

The Internet Research Group sets the tab even higher - at $6 billion for both services and content-distribution equipment by 2004. Most of that will be grabbed by content delivery networks (CDNs), with the remainder going to ISPs and Web hosting companies that operate content sites.

Cache values Of course, CDNs were not the first to bring content to the edge of the network where it could be more cheaply and reliably sent along to end users. Caching was adapted from the semiconductor industry, where chip makers had been putting a small amount of memory on microprocessors to store frequently requested data bits and speed operation.

Caching powerhouse Inktomi, for example, introduced its Traffic Server product in 1998. Running on a server supplied by the ISP or Web host, Traffic Server monitors the network to identify the most frequently requested objects on popular Web pages such as corporate logos and activity keys. It then stores them on a server closer to the end user. Traffic Server now includes application program interfaces that enable the addition of third-party functions such as network security or streaming multimedia.

Inktomi sells its products to about 200 customers now, including content providers and delivery intermediaries such as Adero, Digital Island and Exodus Communications. America On-line is its largest caching customer by far. Other caching vendors also have entered the market, including CacheFlow, InfoLibria and NetAppliances.

Where caching vendors sell software or a software-hardware package, CDNs sell a service. While they may maintain a string of data centers for network management and content processing, their business is essentially network-agnostic. Akamai, for example, locates 4400 content servers in the points of presence and on the backbones of numerous separate service ISPs and interexchange carriers. Akamai content customers run a program that selectively "Akamaizes" portions of their Web content. This process - invisible to the end user - takes the URL that normally directs a browser request to the origin server housing the content and converts it to an "ARL," the designated address for an Akamai server.

When a user requests content from one of Akamai's 400 customers, including the Yahoo Web site, Akamai's software determines which server is best suited for that user and sends the Akamaized content, which then is seamlessly incorporated with the other objects downloaded by the origin server.

CDNs originally enjoyed a few advantages over typical caching solutions. For example, they usually draw on broader information in making routing decisions. The geographically closest server may not be the best choice for delivering content because of unfavorable network conditions; CDN monitoring software takes those conditions into account. Then, by not being limited to a single access network or backbone, CDNs can boast a wider reach than caching solutions that may only extend to a single Web host and its limited set of partners.

As content delivery has grown in importance, many Web hosts have built their caching products into value-added services for their hosting customers. Companies such as Digex, Exodus and Genuity have seen the coming commoditization of access and colocation - their core products - and have made a conscious decision to move into secondary services - such as security, storage, backup and redundancy, network monitoring and management and, importantly, caching and other content delivery services.

Hosts with the most - to lose It's the last-named shift that puts the big Web hosts in apparent conflict with the new generation of pure-play content delivery networks. Content delivery is a zero-sum game, and it's stacked against the Web hosts, according to some analysts.

"Netsourcers are having the cream skimmed off their revenue pie by the likes of Akamai," says Alex Benik, an analyst with The Yankee Group. He offers the example of a large Web content provider that pays its Web host $15,000 per month: $5000 for network connectivity, $5000 for rack space and $5000 for additional services. If the content company then chooses to have its content delivered locally by a specialist such as Akamai, the load on the origin server will be reduced - shrinking the network fees the content provider owes to the Web host. In addition, a content provider that requires less server processing power will need less rack space - creating another revenue hit for the Web host.

CDNs have a more compelling sales story than cache vendors for the content providers that are their real customers. Caching has four primary value propositions, according to Peter Christy, vice president of research services at Internet Re-search Group:

- Bandwidth multiplication. If a cache can catch half the requests for the content it stores, a provider has doubled effective upstream capacity.

- Performance enhancement . Proximity of content to the user can improve page request times by a factor of two to five.

- Flash crowd control . Keep those Starr reports loading at peak times.

- Bandwidth arbitrage . Bandwidth near the center of the Internet is relatively expensive - around $800 per Mb/s per month.Why not serve it up near the edge, where the bandwidth is cheap or - in the case of a corporate LAN - even free?

Under this model, Christy says, the only value cache vendors can really monetize is the first, that of bandwidth multiplication - and that matters more to ISPs than to the content providers whose Web sites reside on their networks."Although Inktomi and others have done a lot to promote the concept of caches as the platform for value-added - additional revenue - subscriber services, the potential is still unclear," Christy says. "What we see is the Internet access price eventually going to zero, and unfortunately, 150% of zero is still zero."

By contrast, CDNs implement a network of caching servers, add some enhancements and then sell a service to the content provider, which cares less about that first benefit of bandwidth savings but cares a lot about the other three. A Web site that is inaccessible or that cannot scale to meet sudden, foreseen traffic spikes will earn a dot-com a spanking on the revenue bottom line, and usually another in swift succession from investors.

Caching also gets between a Web site and its visitors, preventing content providers from knowing exactly how many times a page is being delivered. These traffic figures can be important for calculating advertising revenues.

Way cool content One proof of the relative strength ofthe CDN value story is the number of Web hosting companies buying into it by reselling Akamai's services to their hosting customers. The company received 6% of its revenue in the first quarter from resale agreements with Digex, Global Center, IBM, NaviSite and other partners. These companies have bitten the bullet and decided that customers who don't get the content services they want will take their Web hosting business elsewhere.

Akamai, whose name means "way cool" in Hawaiian, has servers located in 160 global networks and recently announced a new partner program that may put it into many more. Analysts estimate that Akamai has the lion's share of the content delivery market - perhaps up to 60%.

To date, the CDN has been working mostly with large operators - Cable & Wireless, Digex, Genuity, GlobalCenter, IBM, KPNQwest, NaviSite and PSINet. These sellers commit to buying at least 6 Mb/s of content distribution service every month, at a list price of $6000. The resellers then resold that content distribution in smaller increments to the content providers they host.

The Global Content Exchange Initiative makes it feasible for relatively small ISPs to resell Akamai services by allowing them to buy 1 Mb/s of content distribution per month. Comstar.net, EarthStation Netaxs and Fastnet are the first small ISPs to sign up for the program.

"A lot of smaller ISPs that are part of our Akamai accelerated network program have relationships with small and medium[-sized] content providers," Seelig says. "This is a relationship business. If a content provider really trusts his ISP and wants to do his local hosting through him, then we want to help bring him a really powerful content delivery service. That's the classic Internet story of small content providers that have grown from being small to medium or even very large businesses. GlobalCenter's been with Yahoo since they were a couple of guys trying to do a little directory service."

Akamai also has expanded its scope beyond "simple" content delivery. For example, in June the company announced a global load balancing service called FirstPoint. "We had this powerful technology to make real-time decisions about which of 4200 Akamai servers is best," Seelig says."Then the folks at Yahoo said it would be really useful if we could pare that down to a much smaller number of data centers - use those algorithms on a much smaller set of resources."

In addition, the carrier has begun entering into OEM deals with Web switch makers such as Alteon WebSystems, which will build Akamai software into their networking products.

Last month Akamai signed a deal with Inter-tainer to broadcast movies and video over the Web to PCs and Web TVs on demand. More of such services are in the works, whether from Akamai itself or through third-party providers such as Intertainer, Seelig says.

"We have always wanted our distributed infrastructure to push as many types of bits as possible," Seelig says. "The content delivery offering has evolved over time to support more and more types of content. We went from serving .GIFs and .JPEGs to serving secure file downloads, to Macromedia Flash objects and now streaming."

Take Akamai's competencies at managing hundreds of service provider relationships and mapping users to the right places on the network. The company then installs the servers, ensuring that you have lossless content at the edge, and you have, Seelig says, "a pretty powerful media delivery infrastructure."

"Our business is about a distributed platform," he says. "If we were to go into the hosting business, where things are supposed to be centralized in one place, that would be straying from our roots. But anything that happens best on a distributed system feels very close to what we know how to do."

Networks on the edge Squeezed by services such as Akamai's on one side, Web hosts soon may feel another pinch if facilities-based service providers decide to muscle in on content delivery. That's what AT&T hopes to do. Announcing its new "Ecosystem for Content Delivery" in July, Kathleen Earley, president ofAT&T Data and Internet Services, promised that the company intends to spend $200 million in the next two years to build its own CDN (Figure 2).

Earley promised content providers that the completed network will be able to provide 10 million simultaneous views of broadcast media events. "We're fully committed to providing businesses with the ability to reach as many people over the Internet as national prime-time television reaches today," she says.

By comparison, Digital Island hopes to reach a 7.5 million capacity after a network upgrade; Akamai's abilities are probably comparable.

AT&T's vision is an end-to-end system for content creation, management and distribution in which content providers will use one of four digital media centers in Denver, Los Angeles, New York and Hong Kong to package their analog content into digital and streaming formats. The carrier then will have the capacity to distribute that content throughout its cable and Internet networks, accelerating access using the servers of its Intelligent Content Distribution Service (ICDS). For large corporate customers, AT&T also will be able to multi-cast content over private ATM networks.

The streaming media service is in controlled introduction and will be generally available in the first quarter of 2001.

To build its content delivery network, AT&T is partnering with Alteon, Inktomi, Microsoft, Novell and RealNetworks. Loud-eye Technologies will supply digital media solutions and traffic management applications. Customers already signed include Audible.com, MarketWatch.com, Eastman Kodak and Liberty Livewire, a post-production house owned by AT&T subsidiary Liberty Media Group.

"We look at content distribution as the next generation of traffic on the network," says Rose Klimovich, director of global IP networks for AT&T Data and Internet Services. "So we're leveraging the skills that we have in traffic management, analysis and things of that sort. It fits very nicely into the overall architecture we've got. What we've had to do to introduce content distribution is to put servers and switches at various parts of the network."

AT&T believes it can sell its "Ecosystem for Media" to content providers on the merits of scale, reliability and technical research into the performance issues surrounding things such as streaming media. But the real point is the end-to-end nature of the system and AT&T's ownership of the last mile to many consumers.

"That's really the point of the system - we take content all the way from the provider on one end through the network right up to the end users on the other side," Klimovich says.

Analyst Benik agrees that facilities-based service providers such as AT&T can become important new forces in content delivery."Ultimately, the goal of content delivery services is to improve the end user's experience on a site, and the critical component is owning the edge of the network," he says."Owning a data center is not nearly as important as owning the points of presence."

The question specific to AT&T's CDN is exactly what is it doing at the edge, where purchases of Teleport, TCI and MediaOne and the recent move to assume control of Excite@Home give the carrier a number of significant local assets. ICDS was announced in January when AT&T launched its "Ecosystem for ASPs," but to date, officials have not spec-ified how many servers will be available to replicate and store content at the edge.

These servers are located on the AT&T backbone. To accelerate content to off-network end users, AT&T will rely on private peering deals with other access providers.

"I don't see this as more than a very partial solution to the problems of content delivery," says Jilani Zeribi, principal analyst for Current Analysis. "Without effective peering of content and more detail about execution at the edge, this solution exists mostly on paper right now."

The verdict on peering Some think those content peering deals are the future of content delivery on the Internet: standardized arrangements, including monitoring and settlement, for swapping content among backbones and networks the way data bits are swapped today.

Recent announcements by Inktomi and Cisco Systems promise to accelerate the arrival of those future peering deals to the present - and perhaps give Web hosts a means of staving off the competitive challenge of Akamai and AT&T. Last month, Inktomi announced the formation of Content Bridge, an alliance of Web hosts and service providers - most of them using Traffic Server software - to form a unified platform for content delivery among participants (Figure 3).

"This is the first industry initiative to unite disparate networks," said Inktomi President and CEO David Peterschmidt. "This transforms the Internet from a bit machine into a seamless content delivery platform - a content-centric network of networks."

Under terms of the alliance, Inktomi will supply the core network infrastructure for Content Bridge, while network services provider Adero, in which Inktomi is an investor, will be responsible for updating content across the member networks and for all centralizing billing and settlement of cross-network transactions. Digital Island, Exodus, Genuity, Madge.web, Mirror Image Internet and NetRail are the alliance's first member service providers.

And its first customer is one that could spell success: AOL, Inktomi's marquee client already is running qualifying tests on the EdgeFusion content software developed by Adero, which is set for general release later this fall.

Content providers also will get detailed log information about who has viewed their content, which will help validate their claims to advertisers.

Adero started as a content distribution service competing with the likes of Akamai and Digital Island but will stop selling its services to become the designated Content Bridge clearinghouse.

"We saw we could go down the path of building another proprietary CDN, which is limited to the scope of that individual vendor," says Al Fink, senior vice president and general manager of content management services for Adero. "Now we'll do our core services in serving up control, visibility and performance for content providers on the network, and we'll manage all the revenue sharing position of all the partners of the Content Bridge alliance."

That revenue will be key to maintaining an alliance that, after all, aims at getting collaboration from direct competitors such as Exodus and Genuity, or Digital Island and Mirror Image Internet. Alliance members will resell Content Bridge's services to their content provider customers and keep a share of that payment based on traffic volume. They will remit the balance to Adero, which will distribute a portion to the access providers with caches on the Content Bridge network - again based on the amount of content that has been stored on those caches.

Besides revenue, Content Bridge aims to hold its disparate members by allowing them to concentrate on their higher-end prod-ucts - the things that set them apart from the competition such as authenticated delivery and e-commerce. "More and more content delivery will get commoditized, and you should be evolving into value-added services," Fink says."[Content Bridge] is making a very strict commitment to not being in the value-added service business. We'll be the base enablers of the core technology and will work with the alliance members to define what goes in at the right point in time."

Exodus will rely on Content Bridge to deliver static content, but it feels streamed media and more dynamic content are better delivered from central aggregation points such as those operated by Mirror Image; the two companies have a separate delivery partnership. And secure transactions are better delivered from Exodus' own data centers. "Content percolates up and down the stack," says Niel Robertson, vice president of research for Exodus. "Some things live at the edge; others don't. That's why other content solutions that rely only on the edge are limited. We're trying to maintain real estate all the way from the data center to the edge to be sure we put the right content in the right place."

Robertson's roundabout reference to Akamai points to the CDN's notable absence from Content Bridge's membership roll. The content to be delivered must use HTTP and cannot have been converted to someone else's language. "Our objective is not to cause providers to convert any of their code," Fink said."We want the first true open-standards cross-network content peering platform. Akamai is welcome to join, but their proprietary code will make it difficult for them to fit in."

Speaking prior to the Content Bridge announcement, Akamai's Seelig expressed confidence that no subset of ISPs or hosts - and certainly no single network such as AT&T's - can produce the audience that his company's network-agnostic approach offers content providers. "There's no ISP in the world that has more than 10% of the traffic on the Internet," he says, looking at termination points - which ISP actually owns the final IP addresses to which traffic travels. "You get to 10% of the Web by taking the top 20 ISPs. To get to 95% of the Internet, you need to have relationships with 7000."

Seelig also doubts that content peering will be achieved easily. "It's really hard to do technically," he says. "You need to build a mechanism where you can share logs, share billing records, cross-bill, give real-time reporting off all the caches and servers. If AT&T or UUNet form a content-peering alliance with another Tier 1 network, are they going to allow that kind of monitoring? AT&T's never even been fond of letting outsiders monitor its terrestrial network. And UUNet doesn't like to peer with people - they like to give people transit.

"At 50,000, content peering sounds like this really great idea. At 30,000, it loses a lot of its attraction. And when you get down at ground level, it's got all kinds of hair on it," Selig adds.