The Internet Overhaul

Thursday, June 28, at 12:30 p.m., U.S. Eastern time

Ellen W. Zegura, chairwoman of computer sciences at the Georgia Institute of Technology, is co-chair of a major National Science Foundation project to overhaul the Internet, the Global Environment for Network Innovations.

A transcript of the chat follows.

Josh Fischman (Moderator):
    Is the Internet in trouble? Should it be revamped? Ellen W. Zegura is trying to answer both of these questions. Zegura is chairwoman of the computing science and systems division at the Georgia Institute of Technology. She is also co-chair of the science council at GENI, the National Science Foundation's proposed new testbed for Internet innovation. GENI is a $300-million project to test ideas about network security, mobility, and manageability.

Question from Josh Fischman, Chronicle of Higher Education:
    Most people use the Internet every day, with few problems. Why do you and many of your colleagues say there are signs of decay? What are those signs?

Ellen W. Zegura:
     Let me give you a few examples. The costs associated with network operations are increasing, and currently exceed the costs of equipment. Network operations people are handy persons of the Internet. When they are working hard, that's an indication that fundamental structural changes may be called for.

Question from Kamran Khan, Rice University:
    A major overhaul of the Internet could impact our campuses at various levels (border, internal routers, switches and network management) and has financial consequences. Can you please define “Overhaul of the Internet”? Who will drive this complex undertaking? How consensus will be achieved? Is this going to broaden or narrow the digital divide (have and have not’s) that we are still struggling with in our society.

Ellen W. Zegura:
    At this point, the GENI project is about providing a testbed style facility that would allow researchers to experiment with alternatives. It's too soon to say how the outcome of those research experiments would fold into changes to the current Internet. Certainly the costs of making changes to deployed infrastructure are very high and must factor into ultimate decision making about new technologies.

Question from John, University of Georgia:
    Dr. Zegura, what is your opinion on attempting to regulate the internet? What would be the greatest strength of regulation, and on the converse, what in your opinion would be the downfall?

Ellen W. Zegura:
    I think one of the biggest factors in the success of the current Internet has been its open model, and the associated competition and creativity made possible in that model. Openness does make things like security and reliability much more challenging. I'm eager to see what creative ideas researchers develop for managing the tension between an open network and a managed/secure/reliable network.

Ellen W. Zegura:
    A few of the earlier questions had multiple parts. Let me come back to several of them. One question concerned the digital divide -- what is the relationship between this project and the digital divide. One of the things that the GENI facility could support is experiments with alternative access technologies, such as multi-hop wireless, which could offer different price-performance points for end users. Making lower cost connection options available is at least a step in the right direction for reducing the effects of the digital divide.

Question from Josh Fischman:
    Going back to Kharman Khan's question, could you talk a little about GENI governance? How will scientific questions be defined and addressed? Are there groups or committees already in place and already doing that?

Ellen W. Zegura:
    There are several groups and committees that have worked on GENI and will continue to do so. I am co-chair of the GENI Science Council, a group appointed through an open process to interact with the research community to articulate the research and education opportunities that are possible with a facility. There is also a GENI Project Office being run by BBN that will oversee design of the facility that meets the needs of the research. You can find more information on these groups at www.geni.net. That site will soon include a forum for on-line input and feedback if you'd like more information than you are getting here.

Question from Hank Classe, Delaware State University:
    Where will the push be to upgrade Internet2 in the USA, at the core or the core and stub networks? Will this be inclusive of IPv6? As a stub connection to Internet2, what could be DSU's role in this funding opportunity?

Ellen W. Zegura:
    The relationship between the GENI project and existing infrastructures such as Internet2 is a bit complicated. As I mentioned in response to an earlier question, GENI is a testbed for experimentation. It won't replace the functionality you at DSU get from Internet2. DSU and other stub networks are likely to have methods to get connected to GENI (again, not to replace your INternet connections), where you will find experiments and services that your users may find appealing. Regarding IPv6, GENI will support experiments with network architectures that are far more radical than the relationship between the current (mostly IPv4) Internet and IPv6.

Question from Hank L Classe, Delaware State University:
    IP version 6 was to address many issues which plague the current internet today, inclusive of address space, security, and automatic configuration. Yet its roll out in the USA has been stalled in part to the rather generous allotments held by ARIN. What role do you see IPV6 playing in a redesigned internet?

Ellen W. Zegura:
    I expect to see researchers propose alternative global network architectures that are very different from IPv4 or IPv6. It is too soon to say which alternatives will emerge from experimentation to have a role in a new global network.

Question from Jennifer Ruark, CHE:
    Should the Internet be fixed with possible future uses in mind? What might those be? How can the system be kept nimble, to accommodate change?

Ellen W. Zegura:
    This is a great question. We know from experience that it is nearly impossible to predict what specific future applications will develop. We do know, though, that technologies like mobile wireless are exploding. GENI would allow experiments with network architectures that are designed at the outset to accommodate lots of mobile wireless end users.

Question from Bob Gerdes, Rutgers University:
    Based on conversations at the Internet2 Conference, it seems that GENI project schools would need to be able to take advantage of the New Internet2 Lambda capabilities to be part of the testbed. What kind of bandwidth would be used? 100meg or the full 10gig?

Ellen W. Zegura:
    It's early to talk about this, but I believe there will be multiple ways for schools to get connected into the GENI infrastructure. Some may be host to major attachment points; others may join over conventional Internet paths. GENI should be friendly to connection bandwidths across a range of speeds, without placing an arbitrary minimum.

Question from Josh Fischman:
    What exactly is multi-hop wireless, which you mentioned earlier in connection with alternate access schemes and lower pricing? How would multi-hop help people with lower incomes?

Ellen W. Zegura:
    Most wireless access today takes place between a user's wireless-enabled device (like a laptop) and an access point. There is just one "hop" of wireless, between the laptop and the access point. That works fine when you are near an access point and have the ability to pay the access rate. Multi-hop wireless allows access to occur with multiple wireless hops before you get to an access point. That's helpful for reaching underserved regions that don't have access points deployed very densely.

Question from Simon, Valparaiso University:
    Realizing that nobody can see into the future, how do you plan to balance today's need with the ideal of creating/encouraging a set of technologies that won't cause a repeat of where you see us today in another 30 years ?

Ellen W. Zegura:
    That's tricky, of course. One approach is to make the network highly programmable so that it is as agile as possible to changes in applications and technologies. GENI will support experimenting with that sort of approach. More mundanely, perhaps, there are technology trends in optical, sensors and wireless that we can predict will be of increasing importance. Accommodating and exploiting those will make for interesting alternative designs.

Question from Michaele Lee, smaller grad school:
    Now that we know bigger and faster search engines only help GIGO proliferate, and everyone must have masses of eyecandy (perhaps many of them don't read plain print enough to be able to retain information delivered thusly), do we fix the system or fix the audience?

Ellen W. Zegura:
    I think you're asking about content delivered over networks, including the Internet, and the overload we all tend to experience when faced with massive amounts of information. Dealing with that really falls outside my area of expertise in global networks; I anticipate that the machine learning, artificial intelligence and HCI communities will have much to say about how we deal with these sorts of issues. I don't think the answer is to make the network slower!

Question from Josh Fischman:
    Are there any ideas from any of the 'clean slate' Inernet design projects that might solve some of the current Internet problems?

Ellen W. Zegura:
    Yes. Let me mention a project I've been involved called delay (or disruption) tolerant networking. This project is motivated by the observation that in highly mobile and/or unstable situations, it is common for there to be disconnections in the network. The traditional protocols of the current Internet breakdown when the two ends in a communication are disconnected for any period of time. I've been working with colleagues at Georgia Tech and UMass Amherst on alternatives that allow packets to be stored during periods of disconnection. This simple idea has implications for many of the ways we use networks. There are also some creative ideas for making network management more robust -- Jennifer Rexford at Princeton has nice work in this area. Nick McKeown at Stanford has proposed ideas to eliminate congestion in networks (too many packets in too little bandwidth) which could greatly simplify control. These are just a few examples -- the research community has been generating lots of interesting ideas in this space.

Question from John, University of Georgia:
    Speaking of the machine learning, artificial intelligence and HCI communities, how much of a role do you think they will be able to play in the GENI project?

Ellen W. Zegura:
    I think there will be interesting opportunities for many parts of computer science. There is some work to do on figuring out how to best be inclusive, but the intent is there.

Question from Josh Fischman:
    The notion of a programmable, flexible network isn't really new. Yet there seems to a problem in grafting programmability and flexibility onto the current Internet. What is that problem? Where are the sticking points, in your view?

Ellen W. Zegura:
    The routers used in the current Internet are mostly closed systems that do not allow for external programmability. Plenty of programmability is, of course, available at the end systems, which is part of why we have seen an explosion in peer-to-peer applications such as file downloading. It is tricky to provide programmability in a way that does not threaten stability. The folks involved in designing the GENI infrastructure, such as Larry Peterson at Princeton, have worked hard on ways to provide programmable "slices" that can be isolated from one another. So you get to program in your slice, and I get to program in my slice, and the management system oversees the slices and ensures overall stability. If you make a mistake in your programming, it may affect the people using your slice, but it won't affect me.

Question from Josh Fischman:
    Earlier you mentioned that network operations people are working pretty hard to keep the current Net going. In a more programmable network, won't these people need to work even harder?

Ellen W. Zegura:
    Let me be more clear. GENI (the experimental facility) will allow programmability so that different researchers can experiment with different network designs and ideas. For example, Jennifer Rexford can implement her techniques for simplifying network management in a GENI slice. If these ideas prove useful, one can imagine that eventually a global network (not GENI) might incorporate her simplified network management techniques. So, GENI will have programmability, to allow ideas to be tested. Should a future global network also have programmability? That's a complicated question that contains networking research, policy questions, economic model questions, etc.

Question from Josh Fischman:
    Those clean slate ideas about delay-tolerant network, and congestion reduction, seem really interesting. Where can people learn more about them?

Ellen W. Zegura:
    Most of these projects are described in detail on web pages and (for more detail) in research papers. The people and projects I mentioned are a good starting point, but there is lots of additional great work going on, so don't stop there!

Josh Fischman (Moderator):
    That's about all we have time for today. There's a lot here to think about. Thanks to Ellen for taking the time out of her busy schedule, and of course thanks to all of you for your keen questions. Goodbye.