Needed: self-configuring networks
- By Joab Jackson
- Nov 17, 2004
Col. Timothy Gibson, tech warrior
J. Adam Fenster
Army Col. Timothy Gibson is no stranger to strange ideas. His early career, for instance, included riding not tanks or Humvees, but dune buggies.
It was good training for his present role as a program manager in the Defense Advanced Research Projects Agency's Advanced Technology Office.
Earlier this year, he and other DARPA officials sparked debate by saying the protocols now used on the Internet and other public networks were not adequately serve the military's future network-centric warfare needs.
Before his DARPA assignment, Gibson had long, rich experience in Defense IT. He was director of technology for the Joint Task Force for Computer Network Operations. He managed deployment of the Army's intratheater data network during the first Gulf War. He also was technical director for the commander in chief of U.S. forces in South Korea, and chief of network security and information assurance for the Pacific Command in Honolulu.
Gibson taught at the U.S. Military Academy at West Point, N.Y., where he was commissioned in 1979. He has master's degrees in computer science and history from the University of Kansas and a doctorate in computer science from the University of Maryland-Baltimore County.
GCN associate editor Joab Jackson spoke with Gibson at DARPA headquarters in Arlington, Va.GCN: What's the most urgent military need in networking?
GIBSON: We need secure and reliable systems that can scale to large sizes on their own, particularly in the wireless area.
More than 10 percent of the forces we've deployed for Operation Iraqi Freedom were for network and communications support. We'd like to get that down to 1 percent or 2 percent, but today's network equipment requires large numbers of troops to make sure it is working correctly.
The computing devices we have, particularly the routers, were designed so that once you install them someplace and get them running, you leave them there.
If you decide to move something, you have to reconfigure it. That doesn't meet the military model very well. It'd be much better if a device could configure itself.
If you wanted to move the node, you would just move it, hook it back up to the comm line, and it would automatically reconfigure by itself.GCN: You've called IP a 'probabilistic delivery system.' What do you mean by that, and why is it a shortcoming?
GIBSON: IP works with no guarantees as a best-effort service. If you put a packet on the wire, it will probably get there. If it doesn't get there, the [destination] device will probably know and send an error message back that probably gets delivered. So that's 'probably' cubed, which is not a great delivery record.
The fact that IP is a best-effort service without precedence levels leaves something to be desired. What military commanders really want is a system equivalent to what we used to have with the AUTODIN phone system.
We still have phone systems today that provide different levels of precedence. If someone has to make a higher-precedence call, then a lower-precedence call can be preempted.
A data network should be able to recognize users who have higher priority and give them priority service. So that means you should be able to slow down the data rates offered to other users.
The commercial approach for doing this has been the quality-of-service method, where I have to make a request to the system to give me the bandwidth. That's not the same as precedence.
In the telecommunications sector, quality of service is relatively straightforward. Commercial providers will let their pipes fill up only to about 50 percent. If someone wants another 15 Mbps of traffic, it's easy to offer the additional bandwidth. But military pipes typically use well above 50 percent. The tactical systems are up in the 60, 70 and 80 percent utilization rates.
The commercial quality of service model does not work well when the path itself is fully used.GCN: What ideas are you considering to augment IP?
GIBSON: I'm starting a program called Control Plane. It is going to try to put some control into the network in the same way that a separate signaling plane works in Signaling System 7 [a widely used protocol to carry calls across public-switched telephone networks]. There's a data plane and a signal plane.
In IP, every packet has both data and control data. Hosts just put the data on the wire and have no idea what the path is like between themselves and the destination. But when your system has such marginal path qualities, you'll have problems.
So the intent of the Control Plane is to make the hosts talk to the network infrastructure to find out about the paths between themselves and the destinations.
They should be able to choose the path with the best fit. They should be able to shape their traffic so it goes through the network as efficiently as possible. They should be able to multiplex traffic across multiple paths to dramatically increase throughput or the probability of having the message delivered.
You can apply that same model to the wireless world. When WiMax products come out, they can provide a relatively high-speed data network for users within several kilometers.
When you move into a major metropolitan area, you have these little overlapping clouds of connectivity. The way wireless systems generally work now is to try to attach to the service node with the strongest signal.
But that doesn't provide you or your machine with any information about the best service past the service point. If I have three or four service points, and adequate connectivity to all of them, I would want to use the one with the best service to where I want to go.GCN: So the control plane would be virtual?
GIBSON: We've talked about actually having a separate set of wires. Whether that is a good idea or not is immaterial. The facts of life are, if you come in and say, 'We want to revise the Internet Protocol so much that we will have a separate channel path for the control portion,' there would be revolt like the Internet world has never seen.
Right now, we're looking at putting the control right on top of IP, so it has to be compatible with IPv4 and IPv6.
I think that when we go to all-optical networks, we will have to rework something. Look at how routers function today. When a packet shows up, the router has to make a decision about where it is going.
With an all-optical backbone, packets start at the source and just go zipping through the network infrastructure. There is no router in between; there is an optical switch. But the optical switch doesn't really have to make any decisions.GCN: What are you doing to encourage more development in self-configuration?
GIBSON: We have developed mobile ad hoc networks, or MANETs, that work very well. They provide good connectivity and automatically configure themselves. But they do not scale well to more than 200 devices.
Obviously, that is a problem for the military. A platoon has 35 soldiers. A company has somewhere between 120 and 150 guys or gals, and they are moving around. So you need a system that can scale to thousands of devices and configure itself automatically. We want to develop wireless MANET technology that can scale up like that.GCN: How did you become a DARPA program manager?
GIBSON: I worked with DARPA program managers for several years while I was assigned to Pacific Command and then at the Joint Task Force for Computer Network Operations. Several of the program managers asked me to interview for a job in the Advanced Technology Office. I went through a series of interviews, eventually having one with Dr. Anthony Tether [head of DARPA].
Once DARPA wanted me, I had to convince the Army that having me work here was a good idea. It's the most enjoyable job I've had since I was company commander. The best part is the complete latitude about what you want to do.
If you're not a self-starter, you probably will never get hired by DARPA. Where you put your time and energies is all based on what you want to do.