February 2001

BY JEFF LAWRENCE

CONTENT
The next-generation network will allow the deployment of new services for communicating, entertaining, transacting, information gathering, and education. Not only will the services become more useful, but the overall experience will also become more media-rich. Traditionally, content has been a combination of text and graphics in the form of e-mail and Web pages on the Internet, and voice on the PSTN. E-mail and Web page traffic is bursty in nature and also fairly immune to the delay, jitter, and packet loss that exists in today's Internet. Voice, audio, and video traffic are much more sensitive to delay, jitter, and packet loss.

As the Internet quality of service improves, there will be increasing demand for voice, streaming audio, and video services to enhance the user experience. Text and graphics make very small bandwidth demands on the network compared to streaming audio and video, which will require one to two orders of magnitude greater bandwidth. As the transport of high-bandwidth streaming media becomes available, services such as Internet radio, video-on-demand, interactive TV, video conferencing, and others will become more prevalent. These services will not only be provided in a unicast mode (between a single content source and a single user), but also in a multicast mode (between a single content source and anywhere between a few to millions of users).

ARCHITECTURE
Presenting content to a user depends on the display capabilities of the device, which can range from simple text to full color digital video. Some deployments address the display limitations of simpler devices by using an architectural model in which the content of a Web page is stripped, reduced, and stored on intermediate servers before being displayed.

The format of the information stored on these intermediate servers depends on the type of device it will be displayed on; in fact, many logical sub networks with similar content are necessary to support many different devices. This approach is not very efficient, scalable, or flexible and may result in situations in which information that is available within the network is not actually available to the user. In the future, it is likely that architectural models will favor a single content source that, after its transport through the network, will be transformed into the appropriate format for the device for which it is destined.

The use of firewalls and proxy servers has virtualized the user. It is usually not possible to identify a single user by a persistent source IP address and in fact, multiple user requests from a single user may use different source IP addresses for each request. This has driven the use of cookies and Secure Socket Layer (SSL) identifiers as a means for the content source to ensure its responses to a request are provided in the context of earlier requests from the same user.

Content may be static, dynamic, or streaming. Static content consists of graphics and text. Dynamic content is individualized for a single user by scripts or applets that perform their actions depending on the identity of that user. A single Web page may consist of one or more content elements that, because of their differing nature (i.e., data, executable program, etc.), may be spread across different servers within a single data center, and even between different servers in different data centers. The content from a single Web page is spread across different servers to take advantage of efficiencies in their performance, cost, and storage capacity characteristics. As an example, scripts and applets would require high performance, and streaming media would require significant amounts of low-cost storage. Interestingly, the user is not aware of the distribution and virtualization of the content since it is bound together into one response before it is presented to a device.

In addition to the virtualization of the content, the servers themselves have also been virtualized. Server farms are represented by a destination IP address that is typically intercepted by a load balancer. The load balancer routes the user request to a specific server based on its availability and load. Hiding the servers behind the load balancer allows a data center or hosting site to smoothly add capacity and ensure high availability in the face of server failure or maintenance activities, all with minimal impact to the user.

Content distribution, at it simplest, is about efficiently moving bits and packets around the network at wire speed and changing them as necessary to be displayed on a device. Any content distribution strategy must scale and easily bind together content elements such as text, graphics, voice, audio, and video. Following are a couple of architectural approaches to meet these requirements:

• Load balancing on a network-wide basis would distribute user requests among various data centers or hosting sites that could most quickly provide the user with the content they want. The destination IP address in this case would point to a load balancer, which splices itself into the connection between the user and the ultimate data center or hosting site.
• Another approach is to cache content throughout the network. The content is duplicated at multiple locations, and the destination IP address would point to a content router that would in turn point the request to the closest available content cache.

Common to both of these approaches is using the Domain Name Server (DNS) to identify users, load balancers, and/or content routers. As part of this process the load balancer and content router must have knowledge of the network and server topology, network delays, packet loss, and server loads to find the most efficient path to the content.

As the content is moved through the network there are many applications, including intrusion detection and access control, that must also be able to examine the packet stream from the content source and ensure that it does not violate any restrictions imposed by the user or their device. Media streams are stored in many different formats, and as they are transported through the network they may need to undergo a transformation (also known as transcoding) from one encoding and compression scheme to another, before final delivery to the user device.

The complexity of this problem increases as one content source supplies content simultaneously to many devices with different display characteristics. Also, as the content flows through the network, via either of the preceding distribution models, additional content may be spliced into the original stream to provide some localization of the final content stream. The best example of this would be a national news feed into which local advertising is inserted during commercial breaks.

CONCLUSION
The challenges of distributing content on the network are going to create a new class of challenges that are characterized by wire speed performance, scalability, cost, and flexibility. Content distribution strategies and approaches are in their infancy, and solutions will continue to evolve to meet the demanding requirements imposed by future network services. These new architectural and technological approaches will need to take a broad perspective, and seamlessly integrate and match the disparate computing, packet processing, and communications capabilities of the network infrastructure to truly deliver the solutions that will be needed for the future.

Jeff Lawrence is chief technology officer of Intel's Network Communications Group. Jeff was formerly president & CEO of Trillium Digital Systems, a leading supplier of communications software solutions. 

[ Return To The February 2001 Table Of Contents ]