This was a presentation by a Ph.D. who quit his systems programming job a few years ago and formed his own company to do expert systems for MVS systems programmers. He brought his product, a PC to run it on, and talked about its construction. Pretty interesting stuff here.

His system asks you a lot of nosy questions about various and sundry performance measurements of your system, then prescribes things you can do to improve response time. It was conceived back in his systems programming days, when he spent half of his time doing systems programming, and the other half training new systems programmers! The rule-based system he demonstrated was intended to be used as a training vehicle first, and a real tuning tool next.

He has written a series of about 500 rules that apply to MVS and CICS environments. They execute under a "shell" he wrote in Turbo Prolog. Because his company is a start-up, his choice of inference engine (Prolog) was influenced by the fact that it was CHEAP; there are lots of shells out there which cost an arm-and-a-leg. His effort has cost him about two years so far.

So is it any good? I don't know. It is a little rough still, and is chock full of spelling errors. It does have a very nice set of context-sensitive help. It explains how it arrives at its conclusions. The rules are fairly general and the recommendations it gives are not very specific - they assume you are already able to perform some fairly sophisticated systems programming tasks.

I think I would be more interested in this product if it would take some of the measurements it asks for itself, and then DO some of the systems programming it asks you to do for it.