IBM pulls out its 'Blue Genes'
IBM also has been working for more than a year on a feature called software rejuvenation for Windows servers. Unfortunately, servers using Windows must be restarted periodically because of problems such as memory "leaks"--when computing processes claim memory but don't return it when done.
Consequently, IBM's Windows servers can automatically restart themselves periodically, and IBM has been working to make the feature more sophisticated, predicting when restarts are needed so the server is available as much as possible.
Also tying into eLiza is Blue Gene, a coming IBM supercomputer devoted to the task of figuring out how genes "fold" molecules into gigantic biochemical molecules called proteins. Blue Gene will have so many CPUs that the computer will have to be able to assess when they start or stop working and adjust accordingly.
IBM also is working on lower-level technology, Patrick said. The company already has begun selling memory systems that can keep working even when memory chips fail completely.
All these fixes may seem complicated, but IBM thrives on complexity. Much of the revenue of its large services division comes from helping customers handle onerous chores such as adding new computers to older networks or running customers' systems at IBM for a fee.
There is some risk to having servers managing themselves automatically, where they have enough power to damage themselves as well as heal themselves, Eunice said. But companies with large numbers of servers don't have the luxury of having enough people to control all servers and computing infrastructure, he said.
"Automatic actions have some risks, but the honest-to-God truth is we have no option. If you have 100 servers, it's no trouble. When you have 4,000 or 40,000, you have incredible trouble (managing) by human intervention," Eunice said.
IBM's goal--shared by competitors such as Sun Microsystems, Hewlett-Packard and EMC--is to reduce the difficulties of administering the large servers at the heart of Web operations and corporate networks. There simply aren't enough knowledgeable administrators to go around, particularly as people grow accustomed to having guaranteed access to the Internet and more and more operations depend on the Internet, Patrick said.
Although all big computing companies are working on increased reliability, IBM has decades of research and expects to spend about US$400 million a year on the problem, Eunice said. "The mainstream competition like Dell, EMC, Microsoft and Intel just doesn't have the resources to compare here," he said.
A very small number of computer experts are able to diagnose thorny problems in the most complicated combinations of computing hardware.
"We're trying to capture that knowledge and automate the process," Patrick said. "We can see a real crisis ahead as the expectations go up and the transactions go up."
IBM didn't base the project's name on Eliza, a storied pre-PC program that performed psuedo-psychoanalysis on its users, but on another biological system. The name is a reference to IBM's Deep Blue chess-playing machine, which Wladawsky-Berger said had the intelligence of a lizard--not very smart by some measures, but not bad for a computer.
While the project is ambitious in its scope, IBM has a bigger footprint in the computing industry than any of its competitors.
"I definitely think IBM's the right company to try to attempt to get this kind of technology," Greenberg said. "It's a huge, huge project."
