Darpa Looks Past Ethernet, IP Nets

REDWOOD SHORES, Calif. -- In a handful of high-level programs, the Defense Advanced Research Projects Agency is pushing beyond Ethernet and Internet Protocol to hammer home the message that peer-to-peer networks -- intrinsically more efficient than end-to-end topologies --are the future, and the future is now.

Darpa is pushing toward a world of ultralow-cost, low-power, ad hoc mesh networks. The programs are part of a broad military drive toward ubiquitous computing based on next-generation networks, including RFID and wireless sensor nets.

At the Wireless Ventures conference here last week, Preston Marshall, a program manager in Darpa's Advanced Technology Office, moved the spotlight away from the rise of today's 802.11 networks in the business world, saying in keynote that an improvement of five orders of magnitude is needed. "We get trapped by the vision of Internet Protocol like its some sort of theocracy when in fact there are much better models," he said.

Marshall's contrarian note resonated amid predictions like the one by a venture capitalist that Wi-Fi nets will ultimately surpass even cellular as both race toward the consumer broadband era.

But such networks are grossly inefficient for sending small amounts of data that will be the hallmark of future machine-to-machine networks of embedded devices, Marshall said. An 802.11b network could take as many as 12,480 bits and 57 acknowledgments to send an 80-bit data packet, a 0.65 percent efficiency rating."This is five orders of magnitude from what we can do. This is like selling cars that get 10 inches per gallon," Marshall said. "We need to think as a community how we can get that efficiency up."

Darpa's vision is to create a new kind of peer-to-peer network for "edge-driven computing." Unlike Ethernet, that network will not depend on packets or predefined client/server topologies with guaranteed end-to-end connections. Instead, it will forward data one hop at a time over a distributed network of autonomous nodes using new and more reliable and efficient schemes.

Darpa is doing its part in at least four major programs now under way. One program is developing so-called connectionless networks using new physical layer chips and protocols that could reduce the energy requirements for communications three-hundredfold. The program includes experiments using silicon-on-sapphire technology to reduce energy requirements in oscillators and mixers.

Another program aims to create systems that automatically scan the airwaves in real-time for available spectrum in which it can set up and tear down ad hoc networks in hundreds or even tens of milliseconds. The project includes development of a machine-readable policy language as the basis for its software.

"This opens up hundreds of megahertz of new spectrum without new licensing," said Marshall, adding that the Darpa program is working in conjunction with the Federal Communications Commission's efforts in so-called cognitive radio.A parallel project is exploring delay-tolerant networks based on work at NASA. In this scheme individual nodes collaborate to form complex relationships forwarding data reliably but without knowledge of the overall topology of the network.