% fortune -ae paul murphy

Advice to a former client

It's nice to hear from you again - and as you can no doubt guess I'm not terribly surprised about what happened with Carson.

On the specifics of your question about which Unix to use, however, I have three comments:

1. first, the Unix flavor you go with doesn't really matter much.

   I've used HP-UX 11.1/3, AIX 5.2/3, and Solaris 8/10 fairly extensively in production environments like yours. All of them work - as does Linux although I've only used that in installations much smaller than yours.

   There are software differences - Linux is a little lightweight and a pig outside its native x86 environment; AIX always strikes me as an eighties Unix with glue-ons designed by Dilbert's boss (think Jamieson) ; HP-UX often seems counter-intuitive to people not used to it; and Solaris on any hardware can be a pain to get working just right - but buy the right people the gear they need and all four can be made to work reliably for you.

2. second, hardware is becoming a much more important strategy driver than it used to be. When I worked with you we just picked the biggest boxes we could afford, but now we have the weirdness that hardware costs have gone down dramatically but strategic hardware risks have gone up.

   x86 servers are still just PCs: offering a good bang for the buck in low security applications but well inside the performance, reliability, and general quality-of-engineering envelope you get with IBM's Power, Sun's SPARC, and even HP's Itanium boxes - although the latter are becoming more and more like their Compaq x86 products.

   The key strategic issue, however, is that software progress is becoming more and more tied to hardware change. In that context, HP's problems with the "Itanic" speak for themselves - technically the Itanium is a great machine, but in practice the software needed to make it a success just isn't there, and I think there's virtually no chance it ever will be. IBM is, I believe, behind in getting its Cell software out the door and the whole PowerRISC transition seems as hopelessly mired in internal and customer politics as its Future System transition was in 1970 -and you know how that turned out.

   My own favorite is Sun's "coolthreads" revolution. This stuff works now, the first products are out the door and exceeding expectations - and while there are big gains to be had in software designed specifically for it, ten year old binaries from Solaris 2.5.1 continue to work and easily accessible optimizations are showing 30-70% throughput improvements over straight recompiles.

3. and finally: remember that it's far more about people than anything else. When Burton came in he had a pretty clear Windows agenda -it's how he made his bones, and it's all he knew then or probably knows now. The next guy, I forget his name, couldn't make it work - well duh, nobody can: if faith is the willed suspension of disbelief, enterprise scale windows is the paid version; a classic emperor's new clothes con - and Carson? well, nice suit, but not much in it.

   The general lesson, however, is that you have to match people to technology: a computer is not just a computer when it comes to management. If the guy you bring in knows and loves HP-UX - buy him HP boxes because he'll do better with those than he will with Sun boxes - and the same goes for all the other brands and combinations.

   Bottom line: Unix is Unix, but think about the future of the hardware most closely tied to whatever you buy into - and remember that success is ultimately dependent on your people. If you decide to go with Solaris on UltraSPARC (and I would) hire a Sun guy - but if the right guy for you happens to be an AIX believer, question his sanity (and yours) but buy from IBM.

(Please note: the names used aren't real, the advice is.)