Dave Clark: Congestion is at Interconnections, Not ISP Core

David D. Clark

Dave Clark of MIT is in the Internet Hall of Fame for his pioneering work. Recently, he has been examining major ISP networks to see where the bottlenecks are. His findings may surprise some people given all the talk of congestion and problems. “By and large the core of the internet is not congested. I don’t think we have a widespread congestion problem.” Networks large enough to have their own fiber backbone can easily add more capacity when needed. They want their customers to have a good experience and make sure their networks are robust.

So where do the well-publicized video slowdowns occur? Dave explains, “Where you see the congestion is at the interconnection points. I think congestion there really is a signal of an ongoing business arrangement that hasn’t been fully resolved.” There is an ongoing public debate in the U.S. and France in the Net Neutrality discussion over whether “sender pays” is a good system. Until that’s resolved, I expect more issues like this to come up. Traditionally, networks “peer” at their edges, splitting the actual cost of the interconnection point. The actual upgrade is usually accomplished very quickly at the peering point. When Netflix agreed to pay the carriers to accept the traffic, most of the problems were resolved in a few weeks.

These are important issues because congestion and the cost of minimizing it is central to the “Net Neutrality” debate. So I’m glad to report an engineer’s opinion from this webinar led by Marconi Fellow John Cioffi. Cioffi deferred to Clark’s expertise on the core of the network. John went on to discuss how today’s vectored DSL and gigabit WiFi can raise the effective speed of home Internet to hundreds of megabits and more.

David Young, a Verizon engineer specializing in policy, made similar observations. In a blog on the Verizon site, he calls congestion a “myth” http://publicpolicy.verizon.com/blog/entry/why-is-netflix-buffering-dispelling-the-congestion-myth

Why is Netflix Buffering? Dispelling the Congestion Myth
Review Shows No Congestion on Verizon Network

After receiving the letter, our network operations team studied the network connection for this customer for the week preceding the date that he emailed us. They measured the utilization – or the percentage of total capacity used – at every link in the Verizon network – from the customer to the edge of our network, where we receive Netflix traffic – to determine where, if at all, congestion was occurring.

This review confirmed again what I’ve explained before (here and here): there was no congestion anywhere within the Verizon network. There was, however, congestion at the interconnection link to the edge of our network (the border router) used by the transit providers chosen by Netflix to deliver video traffic to Verizon’s network.

Here’s Dave’s bio from the Internet Hall of Fame and a partial transcript of the webinar.

David Clark is a senior research scientist at the MIT Computer Science and Artificial Intelligence Laboratory, where he has worked since receiving his PhD there in 1973. He joined the Internet project in the mid-1970’s, and implemented the Internet protocols for the Multics system, the Xerox PARC ALTO, and the IBM PC. From 1981-1989 he acted as chief protocol architect for the Internet, and chaired the Internet Activities Board. During the 1990’s he worked on mechanisms to support QoS on the Internet. He has also studied issues of Internet performance and security. His current research looks at re-definition of the architectural underpinnings of the Internet, and the relation of technology and architecture to economic, societal and policy considerations. He is helping the U.S. National Science Foundation with its Future Internet Architecture program. He is past chairman of the Computer Science and Telecommunications Board of the National Academies, and has contributed to a number of studies on the societal and policy impact of computer communications. He is a fellow of the National Academy of Engineering and the American Academy of Arts and Sciences.

See more at: http://www.internethalloffame.org/inductees/david-clark#sthash.9NIO08Jd.dpuf

Moderator: Dave Clark, some of your latest work has been examining the actual congestion on the internet. What are you finding?

Dave C: What we’re finding that is by and large the core of the internet is not congested. I don’t think we have a widespread congestion problem. We can talk about this a little bit more later if you want, but what we find is that the places where we see congestion are correlated with, I think, widely recognized business issues that are going on today having to do with engineering and payment to deliver high volume content, Netflix and Google.

But if you look at for example a network like Comcast which has 50 plus peering partners and many links to some of those, you’ll find two or three places where there’s congestion. So I think the network itself is in good shape. That’s a US sector comment of course. If you look at links across the ocean, if you look at links at other parts of the world it would be different. But I think the core of the internet today is healthy. Doesn’t mean you’ll always have a perfect experience but we can come back to that.

Moderator: So have we been able to keep up with the volume that’s on the internet, has the core kept up?

Dave C: I think it has. When you look at the inside process, we looked at Time Warner, we looked at Comcast, we looked at Verizon, we looked at a bunch of networks. And some overseas too, British Telecom, Three. You don’t see congestion in the core of their internal network. Where you see the congestion is at the interconnection points. I think congestion there really is a signal of an ongoing business arrangement that hasn’t been fully resolved. But you don’t find congestion inside these core networks.

Moderator: Folks have been handling the speed just fine but the projection is demand is going to continue to grow. The web is going to have a lot more traffic. Are you optimistic about continuing to keep up?

Dave C: I am, I am optimistic. I think there will always be problems, but I actually think that the addition of capacity at the core of the internet is not where we have a problem. The technology is straightforward, it’s a cost issue. The question is do the ISPs have an incentive to invest? We have to be sure that ISP’s are ready to invest. But I think the technology is in place.