Computing on a geodesic dome

by John Boyd

But such complaints don't impress Bill Chute, director of business technology at Salomon Smith Barney Japan (SSBJ) the Japanese arm of the global securities giant. "I alternate between puzzlement and anger when I read reports on how people have gone from mainframes to distributed systems, and the new system costs them twice as much to run," says Chute.

His experience has been just the opposite. In 1990 the company ran on a mainframe, and several minicomputers, workstations, Macintoshes and PCs. Today, SSBJ retains its mainframe for corporate data storage, and has pared the rest down to a thousand workstations and a thousand PCs for its 700 employees. According to Chute, "Our computing costs right now in real dollars, in number of personnel, in inflation-adjusted terms--in every measure I can come up with--is half of what we were spending on computing in 1990. And of course the applications are more complex."

Chute has accomplished this apparent miracle by adhering strictly to four simple guidelines: distribute, automate, simplify and replicate.

Distributed computing lets him flexibly allocate computing power to where it can best be used. He also benefits from the variable costs of scalable technology, while avoiding the high fixed-costs associated with centralized mainframes.

"I can make an investment of \300,000 and boost my computing system," says Chute. "Whereas, if I have to upgrade my mainframe, it's a few million yen each time I need extra memory or disk space; mainframe resources don't scale as well." He also regards PCs, workstations and networking equipment as commodities, so he budgets to replace them every three years, in order to take advantage of more powerful equipment that is continually being introduced.

Automation refers to having computers--rather than people--monitor computers. When a monitoring computer finds a problem in another machine, it immediately informs the IT staff with an audible warning and by e-mail.

SSBJ employs two modes of monitoring: real-time and periodic. In real-time monitoring, systems are watching each other for any faults, such as application errors or a lack of response. Periodic monitoring is more proactive, and involves measuring any deviation from standards and performance. "I get a computer-generated report on my desk each morning, which tells me how all of my computers have done over the past 24 hours," explains Chute. "All the standard behavior is filtered out. Instead, I get to see any patterns of misbehavior, any decay as a result of entropy in the systems."

And because this is a predictive process, SSMB is able to take advantage of vendors' pre-failure warranties. "So I don't have to wait for a disk drive to go down, or for a fatal memory error to occur," he says. "If I detect an oncoming fault, anything out of spec, I can swap the part out at little cost."

Given that he's dealing with thousands of computers, Chute's favorite mantra is "Simplify!" He strips away the frills on machines, insisting they must perform according to established benchmarks. Platforms and user interfaces to Windows NT for general computing, Unix for the mission critical stuff, and the mainframe for large data storage. As for networking protocols, everything has been standardized to IP (Internet Protocol). Concerning Web standards, he is more or less forced into using what's out there, including CGI, Perl, Javascript and dynamic HTML, but he longs for the day when Java might become sophisticated enough to cover all his requirements.

Replication is the automated cookie cutter approach to implementing standards, such as when configuring machines dynamically over the network. "This is the stage where we put back any frills but we do it through mass customization," says Chute. "We don't tweak one machine for one user; if a feature is worthwhile and we can support it, we give it to all users. This allows for predictable behavior from all of our machines."

The result is that Chute can say without hyperbole, "as the numbers of computers has increased, the stability of the overall system has gone up because we have been following the rules of distribute, automate, simplify and replicate. It works rather like Buckminster Fuller's geodesic dome, which grows stronger and cheaper per unit volume, as the dome gets bigger."

John is trying to create his own geodesic dome, but doesn't know what it's supposed to look like. If you have any useful suggestions, let him know at boyd@gol.com.

Back to the table of contents