If all goes according to plan, Haiyong Xie's work developing experimental peer-to-peer could eliminate many of the headaches that P2P systems have traditionally caused ISPs. Last week, Xie helped run the new P2P technology through a field test that was conducted in conjunction with Verizon, P2P software developer Pando Networks and his fellow researchers at Yale. The test was sponsored by the P4P Working Group, an industry organization sponsored by the Distributed Computing Industry Association, whose mission is to bring ISPs, P2P software distributors and technology researchers together to create a set of practices designed to optimize P2P content distribution. Network World senior writer Brad Reed talks with Xie, who is currently wrapping up his Ph.D. dissertation on the P2P experiment at Yale, about how this new P2P technology could make life easier for both P2P users and ISPs.

How did you get involved in helping develop this technology?

The original idea was proposed by me and my advisor [Yang Richard Yang, an associate professor of computer science and electrical engineering at Yale] back in 2006 while I was a Ph.D. student at Yale. We didn't actually get in touch with Verizon and Pando until May 2007, when I was presenting my research at Polytechnic University in New York City. Afterward, I met [Pando Networks CTO] Laird Popkin. We talked about my ideas, and we decided that we should start a working group that would make P2P development companies work together collaboratively. It was late last year that we decided to do some field tests. All of the technologies that we tested were developed a long time ago, but it wasn't until recently that we had the opportunity to validate those ideas and determine if they were really feasible.

What resources did Pando and Verizon provide you?

Pando provided us with their P2P software and their servers, while Verizon provided us with a network topology map. During the field test, Pando implemented an appTracker, which is a server used to communicate with the Pando network and the iTracker. The iTracker is a server that utilizes the information in the network topography map to analyze traffic patterns, among other things.

So tell me a little bit about how P4P technology works? How does it go about finding the most efficient routes to deliver content?

Previous P2P networks used random peering technologies that were almost completely network-agnostic. With this new technology, we provide an iTracker that takes the network topology map and provides suggestions for people within the network to become network-aware. For example, if a link in a particular P2P network is highly utilized at a certain time, the iTracker can let a person know that they should not be using that particular link for P2P transfers.

Essentially, the appTracker server contacts the iTracker and asks the iTracker to give it some guidelines, which are like a peering matrix, to determine which cloud of P2P clients should peer with other clouds of clients. Suppose we have clients trying to peer with one another in three different cities – New York, Boston and Washington, D.C. After analyzing the data provided by the network topology map, then the iTracker may tell the appTracker that the New York peers should connect to DC peers with probability of 0.9, and that they should connect to the Boston peers at a probability of 0.1. In other words, it would be optimal for them to peer with the DC users 90% of the time, and with the Boston users 10% of the time.

It should be noted that when we use this technology, we do not modify Pando's network in any way. We only provide it with guidelines, with some suggestions for who should be peering with whom. We do not modify existing network infrastructure.

Let's talk some specifics. Verizon says that this particular P2P technology will create a 60% improvement in performance for network users, as well as a 50% reduction in network costs. Can you give me a specific example of a test you ran that demonstrated these improvements?

In our field test, we used a video file where all of the clients tried to download and share the file with each other. Because of the role of the iTracker, most of the traffic involved in the test became localized, which means that most of the peers in New York get the file from other peers living in same metro area. Now that we know the network information, we can make better decisions and thus dramatically reduce number of hubs used in the transfer. [Note: According to Pando Networks, peers trying to download files using standard P2P technology get roughly 6% of their data from people in their metro area; with P4P, they get 58% of the data downloaded through people in their metro area. Additionally, the field test showed that ISPs using standard P2P technology had an average of 5.5 long-distance links during file transfers, while P4P transfers had an average of 0.89 long-distance links.]

Whitepapers

• The Trend from UNIX to Linux in SAP(r) Data Centers
• Network-Wide Change Management Visibility with Route Analytics
• A Comparative Study of CDN Architectures
• Best Practices for HP Servers and HP Enterprise Virtual Array in a Microsoft Exchange
• REMOTE CONTROL: How to Reduce the Cost, Complexity and Risk of Managing Your Distributed IT Infrastructure
• Best Practices for WAN Optimization with Data Replication

View more LANS & WANS whitepapers

Webcasts

• Branch Office Trends and Best Practices
• Correlating granular network and application visibility
• Migrating to an MPLS-Based Network
• Round Table Webcast with Forrester Research
• Best Practices for Deploying WAN Optimization for Disaster Recovery
• WAN Optimization Editorial Webcast

View more LANS & WANS webcasts

Special Reports

• Ehternet Services: WAN options mature
• Keeping Spam at Bay
• Network World Executive Guide: The Virtualization Equation
• Get More From Your WAN
• WAN Optimization: How to rev up sluggish applications
• Network World Executive Guide: Perfecting Application Performance Management

View more LANS & WANS special reports