🚀 go-pugleaf

Marty has written good postings pointing out the difficulties about
starting/operating/maintaining a "supercomputer business" (I'm starting
to wonder just what that phrase means, however :-))

He points out the seemingly iron-clad assumption that if your product
(a supercomputer, say) has a life cycle of n years, having an R&D
cycle of n*2 years is fiscally impossible, and (more or less) points
to ETA as an example.

First, I won't get into the business of predicting life cycle, either
in terms of costs or product longevity.  Either is a crap shoot, as
witnessed by many real examples.  I can think of MS/DOS, the IBM
370 architecture, and the CDC 6600 as examples of product which have
had unusually long product lifetimes for mysterious reasons.  Point
here is that Marty's conclusion that the product life of the ETA-10
is four years cannot be drawn de jure, reasonable as it may be de
facto.

Second, although I agree in principle with the above n vs. n*2 
assumption, one of my points in postings before was that this applies
only to the first product (viz. the Cray-1).  The second product
has, more likely, (using x here) of n/2<x<n years, because one
learns from their mistakes (viz. the Cray-2 vs. Cray-1, Y-MP
vs. X-MP).  The third and fourth, hopefully, fit into an n/4<x<n/2
scenario (viz. the IBM 370 extensions upon extensions to the basic
architecture).  

Supercomputer startups, like ETA was, may well fall into the n<x<n*2
range for the first product.  This is all the more reason to have
external forces at work (government, etc.) during this phase to help
with the ever-present fiscal issues.  That is, if we as a society
place value on supercomputer startups.  I am unsure myself, now,
after experiencing ETA first-hand.

Marty, you're lucky you never took that offer.

No :-).

Rob