SUG trip report

[Steve Grandi]

What follows are some of my impressions of the Sun Users Group meeting
held at the Fountainbleau "resort" on Miami Beach between December 5 and
7, 1988.  Life is sure tough when one has to go such dreary places for a
meeting... 

Scott McNealy, Sun's President, said the DRAM shortage is incredibly bad
for Sun (allocations are based on last year's sales, but Sun has doubled
sales and machines are now going out with twice the amount of RAM as
before!).  Sun is now the third largest consumer of DRAM in the world
after IBM and DEC.  He said to compare NeXT's machine with the machines
Sun will be shipping in July 1989.  Sun has been shy of announcements
during the last 12 months, but will announce lots of machines during the
next 12 months.  McNealy said "Sun is no longer benign" in the eyes of the
industry and is trying to distance itself from the SPARC chip by the
establishment of the SPARC Council.

Bernie LaCroute, Sun's executive vice president, was much more specific
about product plans.  While Sun is committed to the 680X0 and 80X86
architectures, most of their resources will be devoted to SPARC systems.
During the next 3 to 6 months, expect 3 new SPARC machines: a low cost
desktop model running at 10-13 MIPS and about 2.5 Mflops; a deskside
pedestal system at about the same $20K price point as the existing 4/110
running at 10-15 MIPS and about 3 Mflops and a "server" system at about
the same $40K price point as the 4/260 running at 15-20 MIPS and about 4
Mflops.  Later in 1989, an ECL system running at 35-40 MIPS and about 10
Mflops should be released (since this system requires a completely new
memory and I/O system to do justice to such performance, delays might
happen; but a "taped" prototype is running now in the lab).  For
comparison, the 4/280 is benchmarked by Sun at 10 MIPS and about 1 Mflops.

Also in the 3 to 6 month timescale are two 68030 machines.  A system at
the $30K price point of the 3/260 will feature a 33 MHz 68030 and crank
out about 7 MIPS.  A desktop system at the $10K and under price point of
the 3/50 and 3/60 will feature a 68030 at a lower clock rate and run at
3.5 MIPS.  For comparison, the current 3/60 is rated at 3 MIPS or so, and
the 3/280 is rated at about 5 MIPS.  Also in 1989, a "486i" will be
released running at about 10 MIPS. 

The two low-end desktop machines (a 4/50 SPARC machine and a 5/50 68030
machine?) will feature some interesting features.  To quote LaCroute, they
feature "revolutionary" graphics using new chips that will "redefine"
desktop performance levels.  The numbers he quoted showed an increase from
2K 3D vectors (per second?) to 130K 3D vectors.  I wish I knew what these
numbers mean; also, how about resolution?  The 4/50 and the 5/50 will also
have an expansion bus ("open" of course) capable of supporting expansion
cards, will apparently use PS/2 form factor floppies (which Sun expects to
be running 4 Mb in capacity in 1989) as a peripheral, will have room for
small winchesters in the desktop package (there was a very nice 3.5 inch
100 Mb drive from Fujitsu shown on the exhibit floor, for example) and
will feature "telephone quality sound" capability.  These are systems that
Sun says will stand up quite nicely against the NeXT machine.  Expect to
buy the low end Suns in computer stores before too long. 

LaCroute went on to talk about more nebulous product futures.  Sun has
FDDI running in the lab but is very disappointed in the cost: the system
will have to reduced to a couple of chips before it will be economical for
all servers to feature a FDDI port.  PHIGS and PHIGS+ are definitely the
focus of a lot of work as is improved GKS performance.  24 bit color will
become the new standard.  High end graphics systems will move from 150K 3D
vectors and 20K shaded polygons to .8-1M vectors and 150K shaded polygons.

Finally, Bill Joy (VP of research and development) gave his traditional
closing talk and as usual was opinionated, interesting and quotable.  In
Joy's words, Sun is trying to accomplish an economic restructuring of the
middle part of the computer industry by enabling multiple companies to
build binary compatible machines.  Sun is not afraid of competition in
open systems, says Joy, and he goes on to say that binary compatibility
such as exists within the VAX line or in the PC and PC-clone market is the
arena in which Sun wants to compete. 

Joy's comments about the denizens of the OSF (chiefly DEC, IBM, HP and
Apollo) are that their support of open systems is a smokescreen hiding
their traditional "customer motel" strategy (a customer checks in, but
never checks out...).  Joy's test for OSF open systems is binary
compatible formats: will an executable and data file running on a DEC
system using the MIPS chip run on anybody else's system using MIPS chips?

Joy gave several examples of how Suns are becoming more open: the internal
bus ("as good as being on the CPU card") which will appear on many
systems, SIMM memory (Joy expected SIMMS to be in baggies at Radio Shack
before the DRAM crisis hit; it probably will still happen), X11/NeWS and
PHIGS and the upcoming plans to make the processor portion of systems
modular so a new faster SPARC chip can be inserted without changing the
rest of the system.  Even Sun has its limits, Joy conceded when confronted
with the newly closed boot PROMS on Suns, but said the code in the boot
sequence was far too "crufty" to let out. 

Speaking of SPARC, Joy says Sun expects to stay on the "performance
doubling" curve through a pipelined GaAs implementation running at 250
MIPS in 1991 or so, but doesn't see a way beyond that performance level in
a uniprocessor level so Sun is exploring multiprocessor systems.  Joy says
that floating point and graphics performance should now be bundled into
the doubling curve instead of just MIPS. 

Phase 3 of Unix development (Joy and friends rekernelizing and rewriting
Unix in C++) is now in limbo pending direction from Unix International,
but Joy expects something like it to happen before long.  Joy said he
actually wants to use C++++-= (a subset providing "safety" of a superset
of C++ providing multiple processor and system implementation extensions).

Joy had several pithy comments on the NeXT machine.  Basically he said it
is "wrong, wrong, wrong!"  Aside from being very "pre-announced", he said
that Mach is not a production quality OS, Objective C is a veneer of a
object-oriented language, that the interface builder stuff is done better
in existing Smalltalk systems, that the box is big and ugly, the 68030 is
old technology and the magneto-optical disk doesn't yet work.  Joy
conceded he might be wrong (!), but said that Jobs had $100M to develop
his system with, whereas he has about $300M per year in R&D money... 

Here is the scoop on release dates for SunOS.  SunOS 4.0.1 for the Sun-2,
3 and 4 architectures is out now (79 bug fixes!). SunOS 4.0.1 for the 386i
will be out real soon with bug fixes and performance improvements.  SunOS
4.0.2 out Q1/Q2 1989 with new hardware support (including unannounced
machines!) plus bug fixes (you will have to to ask for this release from
Sun).  SunOS 4.0.3 out Q2/Q3 1989 with bug fixes, performance improvements
and hardware support.  Source will be available and 4.0.3 will be
distributed to everyone.  4.0.3 will be the last release to support
Sun-2s.  The implication is that 4.0.3 is the stable, final release of
4.0. 

SunOS 4.1 will be out around Summer 1989.  4.1 will pass SVVS edition 2
(implying the addition of RFS, TLI, mandatory file and record locking) and
be POSIX compatible (but one would need to turn on some options to meet
these standards).  4.1 also has internationalization features.  In terms
of performance, the kernel has been put on a diet to make more pages
available to applications.  The windowing system "repaint/damage
propagation algorithm" is improved to eliminate thrashing (front windows
get repainted first and windows are done sequentially rather than all at
once; this change will show up first in the 386i 4.0.1 release).  The C
optimizer has been made faster and the default optimization level (cc -O)
for Sun-3s has been upped to level 2.  There will be an option to allocate
/tmp out of swap space (no disk I/O for small files!).  NFS will now do
dynamic retransmission time and buffer size calculation and "mount point
hangup" has been avoided to a greater degree.  Finally, context switch
performance on 3/2X0 and 4/2X0 systems has been vastly improved. 

Functional enhancements included in 4.1 include asynchronous I/O, even
more than 64 file descriptors per process, memory locking of pages
(including processes; one will be able to lock down a process from an
external process as well) and much more use of loadable modules.
HoneyDanBer UUCP, new AWK, a simple programmer's interface to the dynamic
binding system and a parallel make are all included in 4.1. So called
"Ease of Use" features such as the Organizer, the color editor and the
help system now included in the 386i versions of SunOS will be part of 4.1
for all architectures.  Sun will also bundle some "personal productivity"
software with SunOS such as a low-end word processor and a paint/draw
program since "people expect them" (both McNealy and LaCroute emphasized
that Sun wasn't getting into the application business, just that folks
expected such tools in the same way a previous generation expected a text
editor). 

Beyond 4.1, and not counting the S5r4 release, the SunOS guys are working
on better support for large heterogeneous networks (following the lead of
the automounter and such), an expanded storage hierarchy (tertiary storage
on tape or whatever integrated into the backup system), better resource
management (improved scheduling and page replacement algorithms), using
newer facilities such as VM in other pieces of the system (use file
mapping in stdio, for example) and improving the programming environment
by providing improved interfaces and support for lightweight processes,
dynamic linking and distributed applications. 

Sun's version of System V, release 4 (S5r4) will not see release until
well into 1990.  Sun is basically a subcontractor to AT&T (serving as the
protector of the BSD flame as well!) for certain portions of S5r4, but
will not have the final system to port until Q3 1989.  There are some
features of SunOS (lightweight processes, for example) that will have to
included in Sun's release.  Sun will provide "switches" to provide
"SunOSish" or "S5ish"  behavior (but S5 will be the out-of-the-box
default).  Software developers could ease their burden by installing the
S5 compatibility package in SunOS 4.0 and making sure their software will
install with either /usr/ucb or /usr/5bin in the $PATH.  The folks from
Sun involved in setting the specifications for S5r4 talked a good game and
claim that it will neither be a bloated mess (greatest common factor) nor
a stripped system requiring huge add-ons (least common denominator). 

SunOS 4.1 will NOT include X11/NeWS or View 2 as a bundled part of the
system.  Both these packages will be released about the same time as 4.1,
but will be unbundled (available for "distribution costs").  X11/NeWS is a
merged server/window system supporting stock X and stock NeWS (Sun's image
Postscript system) applications in a common window system (implementing
Open Look window frames and such).  X and NeWS windows can overlap and
text can be cut and pasted between them.  Politics decree that multiple
toolkits must be supported, so several will be part of S5r4, all
implementing the Open Look "look and feel".  View2 is an X toolkit
providing SunView, NDE is a NeWS toolkit (with all the Postscript goodies)
and Xt+ is AT&T's X toolkit with an emphasis on widgets.  View2 is a
reporting of SunView to X and "should" be an easy transition for old
applications into the Open Look world. X11/NeWS plus View2 will not be
supported on 4 Mb machines in their 1989 release.  Finally, as a
"transition aid" SunView 1.X will still be supported in SunOS even when
X11/NeWS plus View2/Open Look is integrated. 

One highlight of the conference was a showing of the "final" version of
Pixar's "Tin Toy" video.  The winner of the "Sex, Drugs and UNIX!" button
award goes to the derivative, but still funny "(Safe) Sex, Drugs and
SPARC!"  Best performance by a vendor was by Software Associates and their
suite featuring Hagen Dazs ice cream sundaes and a balcony with a
spectacular view of the beach and ocean.  I visited the suite all 3 days.
Finally, the 1989 Sun Users Group conference will be at the Anaheim Hilton
and Towers between December 6 and 9.  Isn't that where they hold DECUS
meetings? 

Steve Grandi, National Optical Astronomy Observatories, Tucson AZ, 602-325-9228
UUCP: {arizona,decvax,ncar}!noao!grandi  or  uunet!noao.edu!grandi
Internet: grandi at noao.edu             SPAN/HEPNET: 5355::GRANDI or NOAO::GRANDI