The company plans to sell client and server hardware bundled with software that employs its coding techniques. Among its early customers is Japan's Nomura Research Institute, which on Monday will kick off delivery of Digital Fountain servers, integrated with its own software, for Japanese customers looking to distribute content via satellite to virtual private networks.

Analysts generally praised the technology, which in its current incarnation is geared for PC users. But debate persists over how much demand there is, or ultimately will be, for broadcast streaming media.

"The Internet today isn't designed for sending rich media content to a lot of users," said Mahal Mohan, director of product marketing at Digital Fountain. The goal for the startup is to solve the two fundamental bottlenecks to large data transfer over the IP: network congestion and server scalability.

Each of Digital Fountain's two products — one for streaming files and the other for downloading files — comprises a server and software plug-in to transform original content into streams of "meta content" packets. The original content may range from streaming video and audio to such large software files as operating systems, video games or MP3 files.

The architecture employs Luby Transform (LT) code — named for its creator, Michael Luby, the startup's co-founder and chief technology officer and a former professor of computer science at the University of California, Berkeley. A Digital Fountain server with an LT coding engine receives the original content, scans it and generates meta content — elemental information interpreted as mathematical equations — on the fly.

Each piece of meta content constitutes a random subset of the original content. Information about each original packet is spread across a number of meta content packets. Once the client receives a certain percentage of meta content packets, the Digital Fountain client software solves the math equations and fits the original content together.

"It does not matter which packets are received, in what order the packets come or even how many packets are lost," Luby said, as along as enough packets are received to enable the client to perform the requisite calculations. Lost or corrupted packets are thus less of an issue with the Digital Fountain approach, minimizing or eliminating the need to retransmit data or add oversized redundancy to the original content to compensate for dropped packets.

As soon as the packets become available to all clients in the Digital Fountain approach, any number of people, using meta content packets in any sequence, can re-create the original data. The packet stream can be delivered to multiple clients concurrently, at speeds ranging from 30 kbits/second to 1 Mbit/s.

So far, details of the math involved in the LT codes have not been disclosed to the public. Several patents have been granted for the technique, and others are pending.

LT for big jobs

Coding theory has been around for decades. Forward error correction coding, such as the Reed-Solomon algorithm, is implemented in hardware and is designed to protect small chunks of data, detecting and correcting corrupted bits or bit flips. FEC adds redundancy to the original information.

LT coding, by contrast, is intended "to protect a large file — 1 Gbyte or more — and the technology works fast, in software," Luby said. The data overhead required for coding to recover the original data is "as little as 5 percent of the original content," he said.

Digital Fountain's technology is "a very innovative approach to solving content delivery issues," said Michael Hoch, senior analyst for Internet infrastructure at the Aberdeen Group. "There are other strategies, such as server farms and storage-area networks, but they are much larger and more complex."

Christine Perey, president and principal analyst at Perey Research and Consulting (Placerville, Calif.), said she believes Digital Fountain's technology "will improve and stabilize the cost structure for delivery of digital content and experiences for a significant number of companies who seek to distribute their content over the Internet Protocol." Perey described the technological benefits of Digital Fountain servers as "similar to the improvements one would obtain on a well-designed IP multicast network," but Digital Fountain's meta content packets "offer the additional benefit of reliability without sacrificing scalability."

IP multicast lets a large number of people watch the same file or streaming content concurrently without the provider having to store multiple copies of the content. Multicast saves bandwidth for service providers looking to broadcast over the Internet. But the technology has yet to take off. One reason has been a lack of supporting infrastructure. Another is that IP multicast currently cannot accommodate video-on-demand.

By contrast, the Digital Fountain technology plugs into the existing infrastructure. "Getting service providers to deploy Digital Fountain servers leverages what is in place today and is probably a faster route than waiting for those who own the backbone to implement multicast," Perey said. "It will also simplify network design for premium content delivery."

Further, Digital Fountain's approach can enable video-on-demand because meta content packets allow users to jump in at any point of the stream to reconstruct the original content.

Server details

The two Digital Fountain servers slated to debut next week are the Streaming Fountain, priced at $100,000, and Download Fountain, at $40,000. Slated to hit the market in May, the Streaming Fountain server can work in either a multicast or unicast environment. Used for unicast, the server can deliver up to 60,000 narrowband streams at 56 kbits/s, 10,000 VHS-quality streams at 300 kbits/s or 4,000 DVD-quality streams at 700 kbits/s. One Download Fountain server can serve up to 20,000 dial-up users or 5,000 broadband users when used for unicast.

Digital Fountain estimates that service providers would typically need more than 30 conventional streaming servers to deliver DVD-quality content (at 700 Mbits/s) to 4,000 users. Just one Streaming Fountain server would do the same job, the company claims.

The Digital Fountain plug-in client software required for decoding meta content packets consumes about 400 kbytes; the client's memory requirement is less than 20 Mbytes. Meta content packet decoding can be done completely in software, using "a small fraction of the computational power" required by such media decoders as Microsoft's Windows Media or RealNetworks' RealPlayer, Luby said. "We don't interfere with media players," he said.

Digital Fountain's approach is designed for PC clients. Luby noted, however, that the company is working to scale its technology for PDAs and cellular phones. "We think that PDAs have more than sufficient processing power to decode" Digital Fountain meta content packets, he said.

The company has no plans to license its coding technologies. "We are in the business of selling the box," said Luby. "It takes the right imagination to apply the technology in a real product. We also want to keep control over our technology."

Several customers, including the University of Oregon, Sony Corp. and Cisco, in addition to the Nomura Research Institute (NRI), are said to have tested the startup's technology.

"It has worked perfectly," NRI program manager Tom Misaki, who served as a liaison between NRI and Digital Fountain, said of the technology.

NRI expects several customers, including corporate users and network service providers, to install Digital Fountain products this year.

The question is whether those customers will themselves find a market. "There remains some doubt about how many companies are distributing identical content to tens of thousands of desktops concurrently," analyst Perey said. "If streaming content needs to reach large audiences today, it can be prerecorded and pushed in advance to cache servers on the periphery of the network the Akamai model and then even down to PCs or set-top devices with temporary storage. The same caches serve many needs, so the incremental cost is low."

Another market watcher is more optimistic. "The use of live streaming media on Web sites is growing at an unexpected rate, said Greg Howard, principal analyst and founder of the HTRC Group, which has published a "2000 Content Delivery Service Study" of Web site decision-makers. Use of on-demand services is expected to rise appreciably this year from last year's totals, he said.