🚀 go-pugleaf

WARNING: Nothing about uucp or BBSes in here. This shouldn't be here
anymore. Followups, if neccessary, are directed to
ont.personals.whips.and.rubber.chickens where they will recieve
exactly the respect and attention they deserve.

In article <923011366DN5.61R@tanda.isis.org>, Marc Thibault writes:

>     Hugh is solving yesterday's problems. I/O bandwidth _was_ a critical

It's more of a bottleneck today than ever, especially because CPU's
can run at several hundred MHz and RAM and device I/O can't keep up.

>     factor. It isn't any more when you can have a really fast chip,
>     really big cache for both RAM and disk and smart video with

Cacheing RAM on chip helps a little for a single tasking machine
without an OS, but only marginally for anything else. Likewise with
cacheing disk. Without a psychic sentience, cache algorithms can only
use their time making lucky guesses about future use of a resource and
often end up causing a net *loss* in performance.

>     on-board DSP. With today's technology, the biggest bang for the
>     buck comes from adding local-bus memory and CPU cycles. Keep in

You're proving my point. The ISA bus is the bottleneck, so PCs jump
through hoops doing everything they can to avoid having to use it. Add
a special bus for RAM, run all your video processors with ribbon
cables tieing them together over top of the system bus, find a
solution to the disk problem, and pretty soon you're hardly using the
slow bus at all.

>     mind that today's PC is a multi-processor, with a fast CPU and RAM

And the more processors you add, the more bus congestion you get.
*Each* processor is more than enough to clog an ISA bus by itself. The
only solution is to move to multiple busses, and faster ones. Take a
look at the h/w specs for a VAX sometime, or even an Amiga 3000.
There's a reason a VAX or a microchannel machine costs more than a
"killer" 50MHz 486 ISA machine, and a reason people pay that $.

>     on pretty well every I/O board. Big fast bus is good only if you
>     have massive database processing you do a lot, or lots of I/O
>     things going on, as in a multi-function server. This is computer
>     room stuff, not PC.

It's PC stuff for me, and for most people running a multitasking OS
that does a lot of things for them, even when they're only doing one
thing at a time. Sure, you can run a sieve benchmark really fast if
you have a fast CPU, but special purpose applications like that are
the domain of embedded controllers in your toaster to make the toast
come out right (ok, a little hyperbole here :-) not a general purpose
PC. 

Last night I was playing RoboSport (a great multiplayer/multicomputer
game for PC/Mac/Amiga by Maxis, the Sim City, Sim Big Toe, etc.
people) while a Worldblazer was running UUCP full blast on a serial
port, transferring news, while the SCSI bus was busy streaming a tape
to do a backup of one of the hard drives, and also copying OS/2
sources from hard disk to floppy in PC format. Between turns in my
game I was popping into GNU EMACS to send mail, and work down
the perpetual backlog in my inbox, while all of this was going on.

This is not something I do to show off, it's how I work and play every
day. And this is on a dinky 1987 vintage Amiga, that's 5 years old.
Sure, the graphics in the game were a little slower than when the
machine is almost idle. And when memory hungry EMACS has to start
grabbing 16 bit RAM through the old Zorro II bus, it has to catch up
to my typing now and then. But this machine has a 68020 running at
14MHz, and synchronously to some slower busses, at that. It's all the
coprocessors and busses doing their thing the way they were carefully
designed to that makes the whole system work smoothly.

A modern Amiga with a 32 bit Zorro III bus, or a microchannel or EISA
PC with well matched boards and peripherals can do a lot better.

>     The Bus Wars degenerated into sandbox tussles a couple of years

And a couple of years before that, and a couple of years after, and
...

>     ago. They aren't expected to warm up again (if ever) until we
>     start arguing over the 64-bit standards.

There are other ways to go than wider, but that's for a comp.arch
discussion. I'm not tussling in a sandbox here, just having a little
fun evangelizing the ONE TRUE TRUTH (tm) :-) Turn to your favourite
magazine, and pick your favourite OS (no, DOS/Windows doesn't count)
and look at the benchmarks comparing an ISA PC against well balanced
EISA and microchannel machines. Or run you own benchmarks, If you can
go to the work to design a good one. There are things that can make an
EISA or microchannel machine slow, and a lot of the early ones were.
And a "halfway" machine that has an ISA bus, but uses other busses and
brilliant kludges to work around it, will do better than a straight
forward ISA bus box. But if you compare, you'll see that CPU speed is
much overrated in the PC Clone ad hype wars.

-- 
# Hugh D. Gamble  UUCP: hugh@kink.UUCP  Domain: hugh@kink.PhaedraV.On.Ca      %
# AMIX: HGAMBLE  bix: hgamble  voice: +1 416 267 6159                         %
# "My Canada includes all of usenet, even the naughty bits."                  %