O'Reilly NetworkO'Reilly.com spacer
BooksAll ArticlesSafari BookshelfO'Reilly GearFree NewslettersSearch
Linux DevCenter Linux FAQs | Articles
O'Reilly Network: Linux DevCenter

Sponsored by developerWorks

Check out the Open Source and Linux zones -
fast, free, open -
for tools, tips, code, projects, tutorials - need it? get it.


Free Plug-ins:

Download the WinAmp music player

Get the RealPlayer plug-in

Download the Real Jukebox MP3 player


O'Reilly's Peer-to-Peer Summit Synopsis: Transcript


Visit the Open Source Roundtable to listen to this discussion.

Dave Sims: Tim, you had people from peer-to-peer projects like FreeNet, Jabber, and Gnutella there, but you also had traditional companies, people from companies like IBM and Red Hat there. What were you trying to accomplish with the summit? What did you get out of the grouping you brought together?

Tim O'Reilly: Well, there were several things you have to understand about the meeting. The first was this was an exploratory meeting, but the main purpose was really to combat some of the ideas that I've seen floating around in the press. Everybody's hot under the collar about peer-to-peer as a result of Napster's success, and they're putting out a lot of messages that peer-to-peer is about the theft of intellectual property or the breakdown of the idea of intellectual property; there are a lot of scare stories. I look at this technology and I see it as part of a much bigger picture, and so I was trying to make a statement, partly by the people that I invited but also because I wanted to hear them talking to each other. I really saw the people I invited as falling into three or four groups.

First was the traditional peer-to-peer file-sharing applications that are getting so much attention. The second was a group of people who are looking at instant messaging as an application development platform. It's kind of coming out of the fringes, and it has a lot in common with the peer-to-peer tradition, and I wanted to explore that. I also invited people from projects that are involved in the definition of Web services. For example, there was somebody there from Microsoft and from IBM, the UDDI project, because it seems to me that the Web is itself evolving into a new kind of medium that needs to understand the concept of peering between Web servers in a different way. I was also looking to bring in people from device peering. If you look at projects like Jini or Bluetooth, they're really around the concept of devices finding each other, figuring out what they can do, and going from there. One way I like to characterize it is: If you think of instant messaging as a paradigmatic application -- instant message asks the question, "Are you there? Do you want to chat?" Napster asks the question, "Do you have this MP3 file?" Jini asks the question, "Are you a printer and do you know how to do my file?" All these are about people or devices forming in a sense ad hoc networks that are based on capabilities and needs.

Sims: Jon, your column in Byte last Monday gave a couple of different examples of peer-to-peer architecture. I wonder if we could draw up about 10,000 feet aside from the summit and characterize what peer-to-peer is and how it differs from the Web that we're used to using and the way we've grown used to using this client-server environment.

Jon Udell: Well, first I would like to add to that list that Tim just gave one more case, which is Popular Power asking a question, "Do you have compute cycles available for use?" -- which is not conventionally a peer-to-peer application, but it seems to share something in common with all of these other things, which is the notion that in this case the PC is an active partner on the network, and that really is how the Internet always was. But one of the points that came out very strongly at the meeting is that we've lost that capability in recent years for a couple of very specific reasons, one being the emergence of firewalls and NATs (network address translations), and the other being the fact that the vast majority of new users who have come to the Internet are appearing at dynamic IP addresses, so that although TCP/IP itself thinks generally in terms of end-to-end connections among peers, we've overlaid a kind of a client-server architecture on the Internet in what's become an artificial distortion of what it really is.

Clay Shirky is the guy that really pointed this out very forcefully, and I think he's absolutely right. The key breakthrough for ICQ and Napster was the ability to sort of bypass the DNS and bypass the shifting or uncontactable IP address and get directly to the actual machine or the actual person through some other channel. And that seems like a key development that's going to unlock a real pent-up demand for the kinds of peer-to-peer applications that I think otherwise would have emerged just naturally over time.

Sims: Metcalf's Law seems to come into play here. There's so many network servers on the Web that this seems to have the promise of making so many more machines that are really dumb clients servers, if you will, more active participants in the network. Even if the people who own or are carrying those machines aren't even really aware that they're sort of actively contributing to the network.

O'Reilly: Well, that's actually one of the other things that I think was fundamental to the Napster breakthrough. They simply didn't give people a choice of running a client or a server. They just said, "Hey, these things go together, like peanut butter and jelly. Let's basically give people this cool client, and there also happens to be a cool server."

Sims: You pay to play, you have to give up files in order to get files.

O'Reilly: Well, you don't absolutely have to, but the default condition is that you are a server. That's actually a very, very different condition than, for example, the Web went out with, where the default condition for a lot of people was, "You get to be a client," which is the Web as television. It was ironic because when the Web started, it had a lot of this vision of everybody can participate and things like the URL were a great step in that direction. You can literally reference any other document on the Web. You don't need centralization for that. There is a peer-to-peer addressing architecture. But as Jon pointed out earlier, there was a breakdown in that process because at a deeper architectural level where a lot of people, because they had dynamic addresses, weren't out all the time, they couldn't really run their own Web server, and that led to this artificial centralization to some extent. What's really interesting is that in some ways we're starting to say, "How could we re-invest the Web with more of these peer-to-peer characteristics?" as well as looking at new kinds of applications.

One of the things that I think is also really important to think about is the dimension of scope. It's very easy to imagine Napster, for example, as this globally spanning architecture that connects anybody and everybody, but at the end of the day peer-to-peer also has a very powerful element of finding and assembling smaller groups and the natural sort of associations of community. I think that's where there's going to be a very powerful set of applications. How do you build an ad hoc network? If you look at existing peer-to-peer applications, or applications that have elements of peer-to-peer even if they're not sort of purely peer-to-peer architecturally, they do lead to the formation, if you like, of additional interconnections, and the management of those interconnections and the capability to generate those interconnections on your own is a key element of what I think of as the big umbrella of the peer-to-peer.

Pages: 1, 2

Next Pagearrow

Contact UsMedia KitPrivacy PolicyPress NewsJobs @ O'Reilly
Copyright © 2000-2006 O’Reilly Media, Inc. All Rights Reserved.
All trademarks and registered trademarks appearing on the O'Reilly Network are the property of their respective owners.
For problems or assistance with this site, email