The US government, corporate and academic researchers are working on a network that would be able to configure itself, intelligently cache and route data, and allow for fast and reliable sharing of data, all while maintaining military-grade security.

The project is called Knowledge Based Networking and is under development by the US Department of Defense Research Projects Agency (DARPA).

Academic concepts such as artificial intelligence and Tim Berners-Lee's "Semantic Web", combined with technologies such as the Mobile Ad-hoc Network (MANET), cognitive radio, and peer-to-peer networking, would provide the nuts and bolts of such a network. Although the project is intended for soldiers in the field, the resulting advances could trickle down to end users. "Military networks are going to converge as closely as we can to civil technologies," says Preston Marshall, the program manager of DARPA's Advanced Technology Office.

Marshall says that current technology is "dominated by wireless access, not really wireless networking." Instead of using access points to connect wireless devices to a wired network, a Knowledge Based Network would be a decentralized MANET.

"The thing that's fundamentally different in a wireless environment is that the links are fairly unreliable... nodes join and leave the network more or less randomly," says David Passmore, research director for the Burton Group. MANETs would be able to route traffic through this ever-changing set of peers to a networked device or the Internet.

Such networks would have no single point of failure. Conversely, current wireless networks can be can be shut down by removing the access point. Passmore imagines that MANETs might even be formed by computers in moving vehicles but adds that "the routing protocols we have with IP are wholly inadequate to that kind of a situation." More experimentation needs to be done before MANET technology can be standardized and mass-produced, he says, but the Knowledge Based Networking initiative might provide incentive for military contractors to work on the technology.

The ideal MANET would not only choose the best paths for routing packets but would also pick the best radio frequency to use. This would be made possible by so-called "cognitive radio" technology.

Bruce Fette, chief scientist at General Dynamics C4 Systems, explains that cognitive radio "is able to understand the spectrum activity, the network activity and the user activity and select and use the right waveforms, frequencies, and protocols to efficiently support the user and the network." Such a system would avoid interference, either from other wireless networks or from enemies who want to disrupt communications. It would also understand and obey the local regulations on wireless spectrum use. Fette says such a system will be tested in Ireland next year.

Although work is still being done on the artificial intelligence component of this technology, the underlying software-defined radios (SDRs) that allow networks to change signals on the fly are currently being used by the military and others.

By generating the signal in software and then transmitting it with hardware, an SDR can transmit and receive anything from 802.11 wireless network data to proprietary radio communications to television signals. Fette adds that users could upgrade SDR systems through software rather than hardware, making it easy to add new features and types of signals. He says that while SDRs are not widely used, commercial SDR products have been developed. Indeed, the theme for the SDR Forum's 2005 conference was "Software Defined Radio is Available Now," and Fette says that his company has been producing SDRs for the military since 1998. The GNU Software Radio project, which provides free SDR software aimed at hobbyists, was launched in the same year.

A Knowledge Based Network would make decisions about more than wireless spectrum, and Marshall envisions intelligent nodes that could automatically optimize the network. He notes that if a connection spans reliable and unreliable parts of a network, there could be performance issues: if a packet makes it through the reliable region but is dropped in the unreliable part, it would have to be resent through the entire connection. A Knowledge Based Network would automatically break this connection into two smaller connections, one across the reliable region and one across the unreliable region.Then, if data is lost across the unreliable part, it would only need to be re-sent along that region of the network. Marshall estimates that this technique could increase bandwidth tenfold.

Marshall also speaks of a network that would take note of frequently accessed data and save copies on the edge of the network for quick access: "If one soldier needs a piece of map data... the guys around him will need it too." He imagines a dynamic, peer-to-peer version of the caching that companies like Akamai perform. Artificial intelligence could even decide which protocols to use.

Such an intelligent network would not only understand how to move data; it would also be able to understand what the data meant to users. This idea is based on Tim Berners-Lee's concept of the "Semantic Web," which called for Web pages to include machine-readable data in addition to content intended to be read by people. Software "agents" would use this data to understand the meaning of documents instead of simply searching for keywords. In one presentation, Marshall says the Semantic Web would allow users to "access information by content or type rather than by network address."

Although Passmore is all for "intelligent routing of traffic," he is leery of adding too much intelligence to the network. Products such as Cisco's Application Oriented Network recognize different types of traffic and treat it differently, prioritizing some types over others and integrating security features into the network. He fears these systems could impede the deployment of new services and lead to vendor lock-in. "What has made the Internet so successful is because it's a bunch of dumb pipes," he says. "Once you start embedding too much knowledge... you lose all your flexibility... There's 30-plus years of experience that tells you that dumb networks are better than smart networks."

As with all networks, security will likely be an important issue in Knowledge Based Networking. Passmore notes that with MANETs, there is the additional risk of "routing algorithms being controlled or taken over by the enemy or by some hacker," but otherwise, the security risks are "the same issues that we're all familiar with, like denial of service." Marshall says that Knowledge Based Networks will use a technology similar to SSL, adding that "we'll probably keep the security work a little more isolated" from the rest of the project.

The plan for a self-configuring and -managing network holds promise for more than just the Department of Defense. Corporate IT departments could likely benefit from wireless networks that optimize themselves to increase bandwidth and avoid interference from the break room microwave, and Passmore imagines cheap, easy-to-use MANET hardware becoming available to consumers for home use.

Although Knowledge Based Networking could push some major advances in wireless technology along, Marshall admits that the project is not with out its risks: "being DARPA, we could always be wrong, too," he says. "We're allowed."

Top of Page

BACK to STEP News