Newsgroups: comp.sys.next
Path: utzoo!utgpu!judge
From: judge@gpu.utcs.utoronto.ca (Peter Judge)
Subject: Next NeXT
Message-ID: <1990Mar27.205146.2820@gpu.utcs.utoronto.ca>
Organization: UTCS Public Access
Date: Tue, 27 Mar 90 20:51:46 GMT

	Leaving the issues of software aside for the time being (I think
thatUs where NeXT has been putting most of its energies, and rightly
so!) I'd like to add my two bits worth concerning the hardware
platform the next NeXT might offer. There is a lot NeXT could do to
spruce up its MIPS rating without using this week's hottest RISC
chip.

Exploit the advantages of MACH!

	Raw Speed and Multiprocessing
		Like it or not (and I don't) MIPS is a major influence on
workstation purchasers. IUm getting tired of explaining the benefits
of a completely object-oriented development system, rich bundled
software, and so on to skeptics who then dismiss the important issues
with, "yeah but a SparcStation is faster". Besides, IBM's Series 6000
machines with NeXTStep are starting to look good if your application
doesn't need sound (that doesn't include me!), and I think we're all
hoping that NeXTStep on the IBM will be better for NeXT than for IBM,
right?
		I'm sure that NeXT will soon go to the 68040 and that will
settle the score -- for the next six months. Then everyone else will
be on their next generation CPUs (RISC or CISC, I'm not religious).
What to do? How about exploiting MACH's purported ability for
close-coupled multiprocessing! Put a couple of 68040s on the
motherboard, call it 40 MIPS, make it optional to put 4 such boards
in the machine, call that 160 MIPS and then we can get back to the
important issues of software development.
	Why do I have to read that Compaq (!) has multiprocessing
extensions to its operating system when NeXT, which has the ideal
basis for such a system, is still without one?

Laboratory Control:
	The NeXT, with its removable erasable optical disk, DSP and
ability to create virtual instruments is well positioned to replace a
good deal of analog equipment in the laboratory. But, there are a few
omissions which prevent it from taking its rightful place there.

	Realtime
		Incorporate genuine realtime capabilities. System V R4.0 has
some, AIX R3.1 will have them. But where is the pre-emptive kernel,
process locking, guaranteed interrupt response time and other
features necessary to control realtime processes under MACH?

	Nubus Hardware
		The Nubus was a good choice for NeXT. ItUs bus-mastering
features seem well suited for sophisticated hardware and
multiprocessing. How about a fully featured laboratory control board
with two DSP96001s, 64-bit intelligent opto-isolated I/O, multiple
timers, 16-channel 16-bit fast A-D/D-A converters.
	There are a whole pile of scientific applications out there that
are waiting for such a device. The add-on boxes presently available
to enhance the NeXTUs I/O are too single-purpose to be very useful.
	I can't believe that NeXT doesn't have its bus-interface chip
ready yet. Are hardware manufacturers holding off until the software
takes off? Why? How difficult can it be to port the designs from a
Macintosh Nubus board to the NeXT Nubus board. Some re-routing, lots
more form factor and a bus interface chip should make it all
straightforward. So where are the boards without which many
scientific applications controlling the lab must languish?

Other small suggestions
	DSPs
		I second other NetLand suggestions to incorporate the DSP
96001 in the next machine. This chip eases the programming burden so
sharply that itUs worth its cost. It's 40 MFLOP rating takes your
breath away too (is this rating legitimate?) A 'C' compiler would be
another great addition. While weUre talking DSPs, and since we have
Motorola's attention, why not put two on the motherboard and dedicate
one to a Postscript Engine for monochrome and/or colour and the other
to DSP?

	Speedy MODs
		The MODs are great. But when I'm trying to sell its virtues
to prospective clients nobody wants to compare them with the 40-80
Meg tape drive they would otherwise be stuck with, they expect them
to perform like a hard disk! It would be a major improvement if the
MODs sported a transfer rate which supported realtime data
acquisition, e.g., sound to disk, and (of course) if they were double
sided. I would choose increased speed personally.

Peter Judge		(judge@gpu.utcs.utoronto.ca)



-- 
===============================================
judge@gpu.utcs.utoronto.ca	(Peter Judge)
===============================================
