Comment Articles

Autonomic computing changes gear

Rupert Goodwins ZDNet.co.uk

Published: 25 Sep 2002 14:48 BST

Unanswerable questions of our time, number one: if you're so smart, why ain't you rich? And number two: if your new PC's so much better than your old one, how come it don't work proper? Having a gigabyte of fast memory and a two-gigahertz processor is, for a computer, like having a Mensa-grade IQ is for us humans: looks great on paper but in practice just means you get into more trouble faster.

Network together hundreds of these computers and give them complex tasks with huge data sets, and the problems you get managing performance, reliability, security, consistency and scalability multiply out of control. We're hitting the limits.

The answer, say IBM, Sun, HP and many others, is autonomic computing. As the name suggests, this is an analogy of our own bodies' autonomic nervous systems -- the bits that run the mechanisms of staying alive while our higher consciousness gets on with wondering about really important stuff, like what colour to paint the spare room or invading Iraq.

If we had to consciously push food through our gut, synchronise our heart, regulate our temperature and secrete hundreds of varieties of gunk from our glands, we'd die in a quivering heap in about three seconds. Evolution has assigned these tasks to various control loops that do the job fine without us, and the computer companies would like the same for troublesome silicon systems. The dull stuff of keeping the networks going, making sure the right data is available to the right people and the right resources in the right place should be left to the computers themselves. They're good at dull stuff.

This is anything other than simple. Autonomic systems are worthwhile if they save more effort than they require, which means coping with the unexpected events thrown up by the real world. Computer systems are very, very bad at this: biological systems work because of untold millions of years of evolutionary feedback.

IBM isn't daunted: it already has experimental systems that load-balance Web servers by deleting copies of unread site content and duplicating the in-demand stuff, for example. Other researchers have created loose networks of computers that randomly gossip to each other about their neighbours, creating a surprisingly robust distributed picture of what's going on where. From that, other systems can dig out information and resources from anywhere on the network without needing a huge and ungainly replication process and all the management that involves.