Net-centric approach creates a software challenge
Researchers say breadth of Defense operations poses scalability problem
'We think there is huge unexplored territory in fragility, and it is scaring the hell out of us,' DISA's Richard Hale says.
As the Defense Department gears its networks for greater use, its software needs also are changing. Not only will programs need to operate over a network, they must be more nimble, able to be configured quickly to interact with each other and meet the needs of new missions.
At a recent research-oriented workshop in Annapolis, Md., Richard Hale, engineering technical director for the Defense Information Systems Agency, sketched out how DOD wants to run its networks of the future'namely, as a single entity.
The DOD University Research Initiative Workshop gave Defense-funded researchers a chance to get together, compare notes and understand what DOD will be looking for in IT.
Researchers admitted that the network-centric approach will pose new challenges for software developers, especially as more sensor networks and other massively distributed heterogeneous networks come online.
'We have solutions in our labs, but they don't scale up,' said Jeannette Wing of Carnegie Mellon University.A single entity
DOD will consolidate its networks under a single architecture, making use of the Internet wherever possible, Hale said. Today, the agency maintains the Secret IP Router Network for classified communications, the Non-Classified IP Router Network and the public Internet for unclassified information, and the Joint Worldwide Intelligence Communications System for top-secret information. Eventually, the agency wants to merge all of them into a single entity.
Instead of maintaining separate networks for separate missions, it would be simpler and less expensive to use software to separate the layers of service, Hale said.
For instance, state-of-the-art encryption, such as the National Security Agency's emerging High Assurance Internet Protocol Encryption standard, could secure communications, Hale said. The use of multiprotocol label switching, multiservice provisioning and IP Version 6 protocols also would help prioritize data traffic.
Such a merging of networks would allow Defense to better share applications and data, Hale said. The agency has been moving toward applications that can be 'loosely coupled,' allowing agencies to draw on services from each other on an ad hoc basis. To this end, DISA's Net-Centric Enterprise Services will offer a set of core applications, such as collaboration, messaging, storage and authentication, that can be used by all military offices.
Yet, as DISA moves toward this dynamic approach, it is finding that today's software is too brittle to work with these expanding needs, Hale told the audience. Today's software is set up to work one way. Reconfiguration is burdensome.
'We're starting to experience this peculiarity already,' Hale said. 'We think there is huge unexplored territory in fragility, and it is scaring the hell out of us.'
Software researchers responded with some research paths they could pursue.
'We need highly adaptable systems that rapidly reconfigure and continue to operate,' said Stephen Yau, a professor at Arizona State University who has done middleware research for the Office of Naval Research and the National Science Foundation. 'The mission requires modifying, testing and deploying software in hours and days, not in months or years. If it is a mission-critical system, we need adaptation during run time.'Evolving abstractions
Yau suggested developing software agents, or small self-directing programs, to handle real-time duties such as security and quality-of-service levels. 'We wanted things to operate with little user intervention. Software agents are one direction to go,' Yau said. He also suggested developing more lightweight middleware services, which, unlike the middleware produced today for large systems, could be used to link smaller mobile devices.
Researchers also could create more tools with 'higher levels of abstraction,' that will allow developers to build applications more quickly, said Scott Stoller, a computer science professor at the State University of New York at Stony Brook.
Programmers should be able to write programs without getting bogged down in machine-level details, he said. He advocated building development tools that could generate programs from a set of general specifications, or even visual models of how a system would function.
In 2002, ONR gave a Stoller a grant, as part of the $8.4 million Young Investigator Program Awards, to develop software for use in heterogeneous open systems.
Thomas Reps of the University of Wisconsin also said that greater use of specifications and new modeling languages would be important to developing systems more efficiently, though the specifications would have to be updated as a system evolves, which military program managers may not do nearly enough these days.
Last year Reps was part of an advisory team to help the Air Force streamline the development process for writing avionics software for the FA-22 fighter. The code base was about 2 million lines of code, most written in Ada. The software was split into 40 components, which were built by 25 different contractors. Although each component had specifications, they weren't shared and were frequently out of date, which slowed the development process.
'There was no global policy or synchronization, and this led to a pretty sad situation,' Reps said.