Autonomic Networks


What are they?
PDF DownloadComputer networks that manage themselves; sometimes described as “self-managed” or “self-healing.” The term derives from the vertebrate nervous system, which controls involuntary functions like breathing and digestion.

Who came up with them?
Everyone, and no one. Researchers have been investigating autonomic computing for years. Compaq, Intel and Cisco, for example, have all built systems that can test their components and send diagnoses to administrators. The most visible sponsor, however, is probably IBM, which in 2001 launched eLiza, a multibillion-dollar research initiative aimed at building computer systems that can debug and fix themselves even as problems are developing. IBM’s position is that computer networks will soon grow too complex for anyone (human) to properly maintain them.

How do they work?
In the same way that we adjust to new situations—running for the bus, for example, without consciously calculating how many more breaths or heartbeats are needed—a truly autonomic network adapts to changes in the digital environment. By controlling the interactions among processors, researchers are able to develop computer systems that mimic the way the human brain functions. The processors in such systems use algorithms to determine the most efficient and cost- effective way to distribute tasks and store data. Along with software probes and configuration controls, computer systems will be able to monitor, tweak and even repair themselves without requiring technology staff—at least, that’s the goal.

Who’s using them?
Most notably, the government. The U.S. Army and Navy, for example, are each testing the self-healing capabilities of Sun’s Jini infrastructure, and government researchers are exploring autonomic networks for new security platforms. AT&T is securing its internal Web services using autonomics (see story, p. 24). Researchers from the University of Ottawa and Japan have used the principles of biosystems to create TINT, a financial application that decides on its own which stocks to buy and sell. And the first fruits of IBM’s research—which are still in testing—will likely be for traditional applications: intensive supercomputing operations, such as genetics research, drug-interaction modeling and weather forecasting.

How will this change the status quo?
When the network no longer requires so much hands-on tending, technology managers and staff can devote their time to building new business products or addressing the backlog of internal application requests.

What are some issues?
Time to market. IBM’s self-healing research has a multiyear timetable, so no breakthroughs are on the horizon to help administrators struggling today with sprawling networks. Getting competing vendors to cooperate on standards to connect different systems also is a concern.

What’s next?
Terminator? Perhaps. IBM and others are underwriting dozens of academic projects in autonomic computing—to the tune of millions of dollars in grants—over the next three to five years. The efforts will include developing adaptive algorithms for software agents, self-healing servers and artificial intelligence.