25 Oct
1999
25 Oct
'99
6:20 p.m.
The Great Initialed One (spc) states:
Because I develop software. I Edit (Integer), Compile (and an optimizing
compiler really grinds away in core, err, memory), and Run (usually integer).
Since that's what I do, that's what I want a system to do -well-.
I can BLT bits around, too. But it doesn't help me compile data, or
to compute Viterbi syndromes of data, or perform some signal processing
on some data, etc...
"Processing" means it goes through the central "Processing" unit, or
indeed,
through some "channel processor" that modifies data in some way, or at
least uses data to compute some other quantity. Some processor has to
comb over the memory.
Mike Cheponis sez:
Why? Exactly? Well, I want to know just how "powerful" these machines
really were. One way to do that is to actually work on them, and another
way is to compare them to machines that I understand.
For me, it helps put these machines in perspective.
You know, it's funny, I just made this (what seemed to me obvious) observation
about the Dhrystone 2.1 performance, figuring that other people would be
pleased to have some anchor on performance they (like I) could relate to.
But, instead, I've had to change my Asbestos Suit here several times! ;-)
What I'm TRYING to do is to understand the performance of old machines
in terms of new machines; what I don't understand is why so many people
are uptight about this. Heck, we might all even learn something, eh?
Why exactly? it! That's
a very very simple statement of fact, why is this -sooooooo-
difficult for so many on this list?
I don't think it's difficult. Yes, a Pentium is X times faster than a VAX
in integer math. But what I'm curious is to why is that important? Yes
siree, many's the time I wanted my computer to nothing but adding huge lists
of integers as fast as possible. The more the better. I don't actually
care about anything else than adding lists of integers.
Do I sense a wee bit of sarcasm here? ;-/
Peter,
you'll note that I said that there are some cases where "channel-to-
channel" I/O occurs, but I call that "Simulfax Shuffle Time" (after
Firesign Theatre) - that is, in general, except for the special case of
file copy for backup, moving data around without processing it is a logical
and architecture error (or at least, inefficiency).
Please define what you mean by ``processing''. I can have the Amiga load
data off the floppy into memory, then have it output that data to the sound
chips, all without the data going through the CPU. I can have the Amiga
load data off the floppy into memory and have it displayed and moved about
with having it go through the CPU. In fact, some high end [Internet] core routers
don't even ``process'' the
packets that they route. Or rather, the CPU in such routers rarely, if
ever, ``process'' such packets. Data comes in one side, goes out the other
in a more or less ``channel to channel I/O'' process.
The important point is "more or less" AND the whole -point- of routers is
that they are, to the extent possible, simulfax shuffle devices. But if
they were -pure- simulfax shufflers, then they'd be replaced with a piece of
wire.
Those routers are re-wrapping the IP when they send it out another port,
decrementing the TTL field, re-computing the IP checksum, and the low-level
(link layer) wrapper. Sure, much of this is h/w assisted.
(And, as I've said, s/w is just h/w that hasn't settled down yet...)
But if you're going to play semantic games with me, for example, saying that
the channel processors are NOT processors, then, fine. But they are,
indeed, processors. So saying that channel processors are just
simulfax shufflers is cheating (since they are processors).
But it's just a definition. Some say simulfax shuffling is processing;
OK, that's fine, as long as we define terms.
(You conveniently clipped my proof why one bus is often better than two,
but what the heck - you must have agreed with me on that! 8-) )
Like all analogies applied to computers, even this
one is misleading.
Yes, even though the single 2MB/sec gun can deliver the same amount of lead
as two 1MB/sec guns, with two guns I can have better targeting, or have more
of a spread.
I don't see the relevance here with your distinction; can you please
elaborate?
-spc (Sure, I can have the CPU read every byte off
the harddrive (IDE)
AND then process it, but I still get better through put by having
the hardware read every byte off the harddrive (most SCSI cards)
and then process it ... )
I like SCSI, too.
-mac