Mike Armstrong is a past president of SHARE, and a heck of a nice guy. His session was well beyond standing-room only, enough so that they had to schedule another session Wednesday night. The topic: is the mainframe a dying art?

I got there early to make sure I could get a seat; I knew this would be a full session. When I got there I was delighted to find myself on the front row right by the overhead projector, and seated between two other SHARE luminaries. Dave Thiewlis and Robert Rannie are two long-time SHARE personalities, and they are known for their, ah, spirited defense of touchy subjects. I was looking forward to an hour of fireworks.

(Alas, it was not to be. Thiewlis and Rannie were quiet as mice. Maybe it was just as well; this was a sobering presentation.)

Mike spent most of the hour presenting several case studies of corporations that have undergone a "downsizing" effort, dumping their mainframes for networks of desktop systems.

From a purely political point of view, downsizing is hard to ignore. Your CEO gets hammered by the media all the time. Imagine your CEO on a commercial airliner, reading the magazine he found stashed in the seat back. Chances are that he'll read an article that talks about the money to be saved by downsizing.

And there's a lot of money to be saved. Even Jack Keuhler (President of IBM) quotes a four-to-five times difference in hardware costs when you compare client-server architecture with mainframe architecture.

Of course, hardware costs don't tell the whole tale. Are companies really being successful doing this? I mean, REALLY? Mike ran through a handful of real installations that called their effort a success. One company took six months to complete a conversion of all their applications from a 4381 running IDMS to a network of PCs.

There are some gotchas. Sara Lee sponsored a client-server "shootout" last year to benchmark the industry's offerings. Five vendors declined to participate outright. Four others did take part. And the winner was... nobody! No vendor demonstrated the ability to bulk load large databases (you could only load a database one transaction at a time). No vendor was able to produce subsecond response time for a five gigabyte database.

Networks have other liabilities. The technology is moving along at a very rapid pace, and stability is not something you can look forward to. The Intel P5 and P6 chips are being worked on simultaneously, and the P6 chip will be here a LOT sooner than you think. Mike Armstrong thinks a 100 gigabyte LAN will be on the market in 1993. Networks are not particularly fault tolerant yet, though they are considerably more reliable these days. Security and integrity are nightmare issues, and no auditor in the world wants to ask "How do I audit this?" Finally, because the technology is a moving target, technical expertise is hard to find and keep.

And portability is as big an issue under client-server as it is in the big bad expensive mainframe world. Bell Laboratories has identified 153 separate versions of Unix!

Still and all, Mike Armstrong thinks that the movement is inevitable. So he gives the following advice.

To executives:

Reevaluate how you do things, how you run the business. Don't implement information systems that allow you to do obsolete things faster; change the way the system works, and use new information systems to support the change.

Invest in your people. The archetypal COBOL coder-beaver won't be able to keep up in a world of upheaval. (Mike doesn't like COBOL much, he says that "COBOL gives your hands something to do while your mind is off doing something else.")