[ih] Intel 4004 vs the IMP

[Toerless Eckert]

On Mon, Nov 15, 2021 at 11:00:53AM -0500, Clem Cole via Internet-history wrote:
> Maybe, and part of my gut reaction is to agree with you, and then I think
> about what Atari, Nintendo and the like did with the 6502.  Someone like
> Dave Hayne would be a better judge I suspect.  He and his peeps did some
> really amazing stuff with those early processors in those days.

https://www.youtube.com/watch?v=BaTjwo1ywcI

Was mandatory viewing in 1995 when it came out, amongst Amiga fans.
Some interesting tidbids. Primarily that its really nice if you have
chip manufacturing in-house and very short turnarounds.

I don't have any numbers handy, but i started using Sun workstation from
the mid-80th on and comparing to "home-computers", and i would say that
during the 80th already the CPU did not matter as much as DMA based I/O
as the core performance factor to 'CPU based routers'. Of course, one
can see the history of packet/dma based i/o optimization almost into the
2000ths, with designs like MPLS where all you could do was really to move
two DMA pointers around (start/length), and i would claim even our current
protocol designs have not catched up with particle-based I/O which did
start to become available in the 90th more widely (inluding even early PCI
if i am not mistaken)..

And late 70th/80th minicomputer designs of course predated better
I/O over microcomputer designs too. Yada yada: just comparing CPU
platforms won't tell us what the core historic improvement points where.

> So I think we all agree, but the time of the full 16-processors like the
> 8086/Z8000/68000, running on semiconductor memory, and IMP like replacement
> was quite feasible.  But having looked at some of the IMP code that has
> leaked out, and again the existence proof of the game controllers made
> using a 6502, I have a general belief that if someone has wanted to do it
> and *had the right motivation*, folks like Dave and co, would have found a
> way.

To stick to my cunter-priority: What where the first device designs
that could multiplex reading from multiple network interface effectively.
Early designs AFAIK where all built on serial-port chips, and those
didn't have packet buffers but just some sized byte buffers, so the CPU
had to had interrupt multiplexing or round-robing polling, which limited
performance. Not sure thre was any better design before the first ethnert
controllers where CPU then actively could read whole packets at a time.
and then DMA if i get the order right.

Cheers
    Toerless

> That motivation is still also an open question.  The game control folks
> were driven to keep costs as low as possible, while offering
> acceptable/good performance.  Something like an IMP and later a router,
> less so.   Performance, I would have expected to be the high order bit and
> cost would have been important, but people would pay for it.   Hey, the
> early AGC's from Cisco were many, many thousands.
> ᐧ
> -- 
> Internet-history mailing list
> Internet-history at elists.isoc.org
> https://elists.isoc.org/mailman/listinfo/internet-history

-- 
---
tte at cs.fau.de