high performance vs. Unix traditions

[David G. Hough on validgh]

The following paper, available in complete form from Mike O'Dell
at moabellcore.com, isn't about numerical issues directly, but it's
easy to see the connections.  Mike was part of the Prisma team that
was trying to make a supercomputer out of a GaAs implementation of
the SPARC architecture running Unix.

.pp
.ls 2
.fo ;From USENIX Proceedings;-%-;Anaheim June 1990
.po +.75i
.sp 5
.ce 99
.nf
\s+4\fB
Putting UNIX on Very Fast Computers

or

What the Speed of Light Means to Your Favorite System Call

\s-2\fP
Michael D. O'Dell
Consultant
\s-2
.fi
.ce 0
.sh 1 "Abstract"
.pp
A computer with a 250 MHz clock and built from leading-edge
technology works in fundamentally different ways compared with
a one-chip CMOS VLSI processor
clocking at less than 50 MHz.
The interactions between a UNIX implementation and its supporting
hardware have always been quite subtle and remain a considerable headache for
those charged with porting the system.
But in addition to the imprecisions of fuzzy functional definitions
for some key system facilities,
the laws of physics conspire to make the marriage of
very fast computers and modern UNIX systems an even more interesting
challenge than it would normally be.
This paper discusses some of the matchmaking necessary to achieve
a matrimonious accommodation.

Sections:

.sh 1 "The Rules of the Game"
.sh 1 "Suggestions on How to Play the Game"
.sh 1 "Great Expectations \- What UNIX Wants from Hardware"
.sh 1 "Great Assumptions \- What Programs Think UNIX Provides"
.sh 1 "Great Consternations \- Life at the Boundary Conditions"
.sh 1 "Comments and Conclusions"