16 September, 2000

Sharing the streaming burden

by Tania Hershman

TEL AVIV, Israel -- Remember seeing the solar eclipse live on the Web? Probably not, since the overloaded servers buckled under the demand.

Streaming company Vtrails has peer-to-peer technology to solve the problem of television-quality broadcasting over the Internet, but it relies heavily on the kindness of strangers.

Vtrails of Herzeliya, Israel, has developed technology called Full Duplex Packet Cascading (FDPC) for audio- and video-streaming content providers who want to multicast, say, Madonna live in concert to millions.

"(Right now) the server has to send the same content to each user individually. This consumes a lot of resources," says Vtrails CTO Nezer Zaidenberg. "I don't know of any online event that was actually viewed by millions, as opposed to television. No one is able to serve millions on the Internet."

Instead of the server having to feed content to each individual, FDPC creates a hierarchical structure on the fly, with surfers arranged in layers or tiers. Only the top tier of users gets content straight from the original site. Everyone else gets this content replicated in real time by a user on the tier above and passed on.

This peer-to-peer interaction is created through an FDPC server, which sits between the original content server and the Internet.

The replication is performed using a client application that users download to their machines.

The tree of users is optimized using a version of an algorithm called the Minimum Spanning Tree Algorithm.

"The algorithm checks the location of the client and the connection speed," Zaidenberg says. "We make sure you get a replicate from somewhere near you, the same ISP or network group." The tree is arranged so that those with faster connections are on top, feeding the slower surfers.

This concept raises several questions. What happens if the person responsible for replicating for you decides they've had enough of Madonna? "If there is a disconnection, the tree rebuilds itself," Zaidenberg says.

Another issue is that FDPC makes use of surfers' own CPU power to actually replicate the content.

This is the kind of peer-to-peer networking -- where power from all the computers connected to the network is being used rather than just the servers -- that Napster and Gnutella employ and that the CTO of Intel's architecture group, Patrick Gelsinger, recently called the "next era" in computing.

However, with FDPC there is one basic difference: "If you look at Napster or Gnutella, there was strong, compelling consumer need. If you put Napster on your machine, you had access to other users' files," says Jay Marathe, head of consulting at Durlacher Research in London. "This case, it is more based on altruism. There is no benefit to you to run it, it is of benefit to other users."

Also skeptical is Roy Friedman, a senior lecturer in distributed systems at the computer science department at the Technion in Haifa, Israel.

"In theory, it looks great. In practice, there are lots of issues, ranging from security, copyright protection and legal to the fact that reduced traffic on the server might mean less advertisement revenue," he says.

Other solutions to the problem of streaming media-content distribution have been based around the caching concept, pioneered by companies such as Akamai. However, this still relies on server capacity.

The future may lie in IP Multicast technology. This is an Internet Engineering Task Force standard that's an extension of Internet protocol, allowing applications to send one copy of digital information not to just one "receiver," as is the case now, but to a group of receivers.

The networking technologies underlying IP Multicast ensure that this doesn't eat as much bandwidth as sending out the same content many times, grouping the receivers together in a "multicast session."

IP Multicast's advocates -- including Cisco Systems, Intel and Real Networks, all members of the IP Multicast Initiative -- hope that IP Multicast will be widely adopted as the Web standard.

"Deployment (of IP Multicast) is increasing rapidly but we aren't there yet," says Dr. Kevin Almeroth, chair of the Internet2 Working Group on Multicast and senior technologist for the IP Multicast Initiative. "I think we are getting to the point where there is enough deployment to justify a serious look into building applications."

Almeroth was unimpressed by the FDPC concept. "Not that it definitely won't work, but there are much better ways of reducing load," he says. "Straightforward stream replication is one, and then there are startups doing much cleverer things."

Almeroth cites Digital Foundation (for whom he is an advisor) and FastForward Networks, acquired by Inktomi.

"These are two different approaches to the same problem of push-based content distribution, both streaming and non-real time and both audio/video and data," Almeroth says.

Vtrails is hoping to prove Almeroth wrong about FDPC. The patent-pending technology is going to be incorporated in the next version of a Vtrails TourBar product, which enables a "master surfer" to take thousands of people on guided tours around the Internet.

"TourBar is a kind of 'proof of concept' (for FDPC)," Zaidenberg says. "It is our big laboratory to see what happens on the real Internet. What happens when people disconnect, etc."

According to Zaidenberg, content providers are already interested, though he wouldn't name any. However, he believes FDPC will revolutionize the Web as we know it.

"I want the BBC, who can't do this right now, to have their own Web TV station," Zaidenberg says. "It will create a whole new market."


Link to the original article: