[net.works] Experience with HP9000/200

@RUTGERS.ARPA:GRISS%hp-hulk.csnet@csnet-relay.arpa (03/27/85)

From: Martin <GRISS%hplabs.csnet@csnet-relay.arpa>

Our LISP and AI research group (some 50 people) at Hewlett-Packard
Laboratories in Palo Alto makes extensive use (some 70 machines) of
various models of the HP9000 Series 200, including the 237. As an HP
employee, I am unable to discuss any unannounced products, but will
describe my group's experience with currently available products that
we use.

On the majority of our machines we run HP-UX 2.1, which is HP's port
of Bell System III (a non-VM Un*x), based on 12 Mhz 68000's.  Our
typical configuration has about 5-7 Mb RAM, a 55Mb or 135Mb local
disk, 2 or 3 displays (color and/or BW) and a network connection to a
larger file-server, laser printer, etc.  These are effective as
personal workstations for our work with LISP (mostly PSL), LISP-based
screen editors, graphics and programming environments, expert systems
and experimental natural language systems.

Personal experience, and the Gabriel LISP benchmarks presented at the
AAAI last year, suggest that these workstations deliver "power"
somewhere between an unloaded VAX 780 and VAX 750, closer to the VAX
780.  Some of our university friends who run Apollo's and SUN's as
well as the HP9000 Series 200 might be able to compare the speeds and
features; my own experience with Apollo's at the University of Utah
about two years ago was that the HP9836 (then running a Pascal
operating system) was faster than the DN300's we had at the time.
Bell System III is of course not fully compatible with 4.2 BSD, so
random C files brought over from the VAXen often need some massaging.

I have less personal experience with the graphics. Our group currently
tends to do simple (color) vector graphics ("slides" and "icons"),
which we output on color plotters.  We have not done much with the
raster/half-tone images output to laser printers common to many SUN
and Apollo uses that I have seen.

We have recently begun to experiment with HP's new Integral PC, which
is a fairly inexpensive, transportable, 68000-based system running a
ROM resident operating system very compatible with HP-UX 2.1 (C object
files and LISP binary files from the 237 typically run on this system
without change). With its built in ThinkJet printer, multi-window
display, 1.5 Mb RAM, and "quick-start" kernel, people in our group are
suddenly getting rather interested in "portable/home Un*x" systems.

If my comments are felt to be useful (and not too biased ...), I will
be happy to report again when I have something new to talk about.

Martin Griss

griss%hplabs@csnet-relay
..!ucbvax!hplabs!griss
-------

jwp@utah-cs.UUCP (John W Peterson) (03/29/85)

> From: Martin <GRISS%hplabs.csnet@csnet-relay.arpa>

> ...Our typical configuration has about 5-7 Mb RAM, a 55Mb or 135Mb local
> disk, 2 or 3 displays (color and/or BW) and a network connection to a
> larger file-server, laser printer, etc.  These are effective as
> personal workstations for our work with LISP (mostly PSL), LISP-based
> screen editors, graphics and programming environments, expert systems
> and experimental natural language systems.

One thing not mentioned here is cost.  A configuration like that at
retail prices would probably cost roughly the same as a Symbolics 3600,
and the 3600 is still a nicer Lisp environment.  (Though word has it
HP is working on these details...).  

For things like typical Unix hacking, machines like Apollo or Sun are
probably more useful than the series 9000's, because they run "pure"
versions of Unix (4.2 on the Sun, *both* 4.2 and System V on the SR9
Apollo).  The higher end Apollos (460, 660, etc) appear to be Vax/780
speed for most numerical work, and much faster for certain applications
(e.g., compiling C programs).

There is one important point however, you can't ignore the reliability
of HP machines.  They don't break, and are VERY cheap to maintain.