Improving Internet reliability

HP has won a large contract from DARPA to improve the reliability of the TCP/IP protocol.

HP has won a multi-million dollar, multi-year contract from the US Defense Advanced Research Projects Agency (DARPA) to develop technologies to improve the performance of mission-critical computer networks used during combat and other vital operations.

HP’s Public Sector organisation collaborated on the winning proposal with HP Labs, the company’s central research facility, and HP’s consulting and integration business unit.

The project seeks to improve the reliability of TCP/IP (Transaction Control Protocol/Internet Protocol) – the basic communication system on which the entire Internet is based.

Because Internet links are heterogeneous, with dramatically different capabilities, today’s Internet frequently suffers from great variations in performance. High-end throughput depends on avoiding poor-performing network paths and rapidly routing around them.

Despite decades of research, legacy routing protocols often fail to detect and adapt quickly to changing network connectivity and behaviour. Also, TCP/IP often does not perform as well over wireless networks because it is not optimized for “lossy” environments, in which some amount of redundant data is lost during transmission.

“Our work for DARPA is aimed at significantly improving the performance of the Internet, making it even more efficient and reliable in critical situations, such as battlefield operations.” said Patrick Scaglia, vice president and director, Internet and ComputingPlatformResearchCenter, HP Labs.

 “If we can successfully create new approaches to the way Internet traffic is detected and routed, we may start seeing the Internet used as the de facto communications and information network in areas where it previously would’ve been thought too risky.”

In one proposed scenario, the new approaches would be used to optimise the ship-to-ship communications of vessels using digital wireless connections and high-earth-orbit satellites – even if, for example, one or more of the satellites were not functioning properly or high seas had created problematic ship movement.

HP and its subcontractors have proposed two new approaches for the project, one based on a new generation of hardware routers and the other on a set of software routers running on computers throughout the network. Both use a network sensing infrastructure jointly developed by HP and PrincetonUniversity that detects communication link failures at a distance and switches transmission routes from trouble spots to lines where traffic is running smoothly.

The new approaches are initially being tested on nodes of PlanetLab, a global overlay network developed by the University of California, Berkeley, HP, Intel Corporation, PrincetonUniversity, the University of Washington and more than 60 universities from around the world.

PlanetLab treats the Internet as a single, large, geographically distributed computer. Each node on the overlay network consists of a commodity IA-32 running the Linux operating system. An intelligent routing engine performs packet forwarding, and maintains primary and backup routes in the event of link failure.

A common decentralized network measurement and monitoring system securely aggregates and propagates measurements collected at both hardware and software monitors. Decentralized sensing enables enhanced traffic engineering and maintenance, and a novel fault diagnostic tool permits active location of link failures.

The network sensing mechanism will incorporate HP Labs’ scalable network proximity estimator, a TCP fault diagnosis tool and an optional hardware-based packet sniffer appliance for high-speed links.

HP has engaged Anagran, GeorgeMasonUniversity, McAfee Research (the technology research division of McAfee), the Universities of California at Berkeley and Santa Barbara, and Princeton to assist in conducting the research.