Last week's newsletter featured the first of a two-part Q&A with U.S. Department of Energy officials responsible for the Energy Sciences Network, one of the fastest IP networks in the world and a pioneer of IP technology. We conclude the Q&A in today's issue.
ESnet is a nationwide IP backbone linking Energy Department's headquarters in Gaithersburg, Md. with more than 40 sites including many national laboratories. ESnet's ring-shaped backbone network operates at 10G bit/sec in the northern states and 2.5G bit/sec in the southern states.
Here are excerpts from my conversation with Mary Ann Scott, ESnet program manager, and Daniel Hitchcock, a senior technical advisor on ESnet, about their experiences with the latest IP technologies.
Carolyn Marsan: What are you debugging right now for the rest of the world?
Daniel Hitchcock: We're starting to debug how you do optical networks and how you provision optical networks flexibly so you can set up Lambdas from point to point.
CM: Where are you doing that?
DH: ESnet and our [test bed] Science UltraNet together are figuring out how to do this. We're looking at how you adjust the interfaces in the optical switches to make this happen. The vendors claim this capability exists, but it's never actually been exercised. The question is can you hook up two or three dozen of these things together and actually make it work all the way end to end.
CM: What is the status of Science UltraNet and what is its relationship to ESnet?
DH: Science UltraNet is just coming up. The last circuits are now in, and the gear is going in to enable experiments. There are ESnet engineers who are part of Science UltraNet so that what we learn there, we can fold into ESnet. Science UltraNet goes from Oak Ridge National Laboratories [in Tennessee] to Chicago to the West Coast, and there are optical tail circuits that go off of it to Seattle. The bandwidth is two 10G-bit/sec wavelengths.
CM: When do you hope to have Science UltraNet up and operational?
Mary Ann Scott: At the end of April or early May.
CM: What two or three things do you want to learn from Science UltraNet?
DH: The dynamic provisioning of Lambdas. Also, how to hook up an optical network at the end to your optical network if they don't have the same kind of gear. We'll be looking at the systems integration issues and interoperability.
CM: What's the timeframe for the test bed?
MAS: Between three and six years, provided that things are going well.
CM: How soon do you think it will be before you learn something practical from Science UltraNet that you can fold back into ESnet?
DH: From bringing it up and making it work, we'll learn lots of things. You find a lot of things that the vendors didn't think anyone would ever try but don't really work in this environment. There are a lot of things at this level that people don't understand, and you never understand until you build one of these things.
CM: ESnet was an early user of IPv6. What can you tell me about your IPv6 traffic and applications now?
MAS: We have IPv6 fully deployed. The first request for address space in IPv6 came from ESnet. Our laboratories at this point don't have IPv6 fully deployed. They might have some equipment that is partially capable, but it is now down into the laboratories at this point.
CM: So IPv6 is a small share of your traffic?
DH: A very small share of our traffic.
CM: Do you have IPv6 development work going on related to Science UltraNet?
DH: We're ready when the applications come up and the end systems want it, but we don't drive that. Our users will wait until there are really good IPv6 applications available. It's hard enough to get TCP/IPv4 applications to run really fast. IPv6 is a whole new level of complexity, and many of our scientists think that getting the stuff they have today to run really fast is more important than investing in IPv6.
CM: When do you think a major transition to IPv6 will happen for ESnet?
DH: I think that's years away. It will be driven by the scientists at the ends of the network. We're set up to run it, and we'll continue to employ the latest stuff. But this depends largely on IPv6 penetration into the end systems. We have to have more end systems that run IPv6 and have really high performance implementations. We need sustained speeds above 2.5G bit/sec. That's going to take awhile.
CM: What's the most promising network research going on in either ESnet or Science UltraNet?
MAS: It has to be the optical provisioning.
DH: And the distributed trust management. What we understand how to do really well is trust in a military-style organization where everyone reports to the same organization. What we understand less well is how to do trust when institutions trust each other to a certain degree, and institutions trust each of their members to a certain degree. How do you compose security in a sensible way so that you let people get into the things they're supposed to get into and you don't let people who haven't had the training operate the accelerator? If someone is fired by CERN, then all of his certificates everywhere else are also revoked because this person is no longer trusted. This problem is hard, and no one understands how to do that. We're doing this work on ESnet because the science community needs it.