[TUHS] Was artifacts, now ethernet
Tom Lyon
pugs78 at gmail.com
Sun Jul 20 04:53:20 AEST 2025
The other bit of this story that I've heard from Andy is that there was
some kind of gentlemen's agreement between the IEEE 802 and ANSI/FDDI
committees - to have Ethernet stick to 10Mb and let FDDI do 100Mb. Seems,
at least in retrospect, to be incredibly stupid.
And then there was HP with 100Base-VG (iirc). Sigh.
You and Andy understood that faster Ethernet was all that was needed.
On Sat, Jul 19, 2025 at 11:17 AM Larry McVoy <lm at mcvoy.com> wrote:
> I'd like to talk to Robert because I'm willing to bet how 100Mb ethernet
> came to be is not well known. Feel free to forward this to him.
>
> Somewhere in the early middle-ish 90s, I was working for Ken Okin in the
> server hardware division, building Sun's first cluster. It was just a
> bunch of small servers behind a modified kalpana ethernet switch (the
> mods were my version of VLANs, I didn't know VLANs existed at that time).
> The Kalpana switch opened my eyes to what ethernet could do and could
> evolve to in the future if we made ethernet faster.
>
> So I wandered over to Sun's networking hardware and asked if they could
> build 100Mb ethernet over copper. I was too stupid to realize that they
> thought I was asking them to signal at 100Mb the same way they signaled
> at 10Mb. Which doesn't work because of crosstalk issues (which I didn't
> know about at the time, I'm more software than hardware). So they told
> it couldn't be done and I went back to SMCC with my tail between my legs.
>
> It's worth noting that I was sitting one office away from avb and we had
> past history. I got him to redesign some memory interconnect because
> I had actual memory latency results from all the current hardware
> (everyone's not just Suns) and I had a pretty good idea of all the
> roadmaps because the processor architects talked to me because they
> loved the micro benchmarks in LMbench because they were tiny and ran
> fast on their simulators. The design he had was gonna suck and make
> us look bad so he stopped the project and designed a lower latency one.
>
> That's a long way of saying that avb had some respect for me.
>
> One day, a company called Crescendo Communications showed up to pitch me
> CDDI which was FDDI over copper. That signaled at 100Mb. As soon as I
> got it, I asked them to wait, went and told avb he needed to hear this.
> Pulled him into the conference room, told them this is Sun employee #1,
> can you do the pitch again.
>
> It's worth noting they did not ask us to sign an NDA.
>
> We're walking back to our offices and avb sort of grins and says something
> like "Are you thinking what I'm thinking?" I said "yup, 100Mb ethernet,
> nobody wants FDDI packets if they could have ethernet packets".
>
> Here is why it is unlikely that anyone knows about this. Andy did
> something very smart, he said this couldn't be a Sun project, it would
> die if it were. Sun had done mmap, vnodes, NFS, RPC, etc, and the rest
> of the industry was sick and tired of chasing Sun. The whole OSF thing
> was basically "everyone but Sun".
>
> Andy said here is what we're gonna do (I did some but it was mostly him
> at this point): we're getting in our cars and we're calling on every
> networking company in the value, we're looking for a high up engineer
> or their lead architect. And all we're gonna say is "did you know that
> you can signal over copper at 100Mb like this? Wouldn't it be nice if
> we got 100Mb ethernet?"
>
> And it worked. It wasn't a Sun project, noone remembered that I had
> anything to do with it, Andy kept a very low profile, and I believe we
> got 100Mb "ethernet" cards trickling out in about 6 months. In quotes
> because it wasn't a standard yet.
>
> The funny thing is I've done a lot of other stuff that people know about,
> but I'm more proud of the fact that I pushed for 100Mb and it actually
> happened, that's a far bigger deal than anything else I've done (and I
> know, I didn't do 100Mb ethernet but I saw it before anyone else did
> and pushed for it and Andy, and to some extent, I made it happen).
>
> I also did a back of the paper napkin design of the ethernet switch that
> Granite built, Andy found me in a bar in San Francisco (or somewhere up
> there) and asked me if I could have a perfect ethernet switch what would
> it look like. But that's a different story and probably not for here.
>
> On Sat, Jul 19, 2025 at 10:02:12AM -0700, Tom Lyon wrote:
> > FWIW, "VLSI Systems" was Andy Bechtolsheim's company that licensed the
> > Stanford Sun designs to many companies. When Sun was started, VLSI
> Systems
> > was rolled into Sun.
> >
> > Sun's first revenue products were 3Mb Ethernet cards which still said
> VLSI
> > Systems on them.
> > BTW, if anyone actually has one of these 3Mb board, Robert Garner would
> > love to get his hands on one. He's working on a detailed history of
> > Ethernet. Robert's probably not on this list, but I can connect folks to
> > him.
> >
> > On Sat, Jul 19, 2025 at 2:22???AM r.stricklin <bear at typewritten.org>
> wrote:
> >
> > > In terms of unequivocally prized, I???d say the following:
> > >
> > > * HP 9050
> > > * IBM 6152 Academic System
> > > * SGI Engineering Sample #3 - multibus CPU & framebuffer - these are
> early
> > > SUN boards from VLSI Systems who, as I understand it, were trying to
> > > commercialize the Stanford system separately from Sun. Clear genetic
> link
> > > with the Sun-1 CPU and bwone, but not identical to them. Nor to the
> 68000
> > > CPU that ultimately shipped with the IRIS 1000/1200 terminals.
> > > * SGI IRIS 1200
> > > * Sun 100U
> > > * Sun 150U
> > >
> > > In terms of wanting to mention because rarely seen, missing critical
> > > parts, and selfishly hoping to maybe shake something loose someday:
> > >
> > > * Ardent Titan - missing its console and primary graphics board
> > > * Dupont Pixel Systems MacBlitz - missing all the software, both for
> the
> > > Macintosh host and the Clipper C300 UNIX system itself
> > > * IBM 9377 Model 90 - actually not missing anything, but I???d quite
> like to
> > > hear that anything related to IX/370 or AIX/370 survived somewhere
> > > * mips RS4230 - The requisite RISC/os v5.01 media did turn up somewhat
> > > recently, thanks to everyone involved in that effort. Now I???m just
> hoping
> > > to eventually stumble over a new enough version of RISCwindows that
> will
> > > support the console framebuffer (v4.11 IIRC)
> > > * Pixar Image Computer - missing host interface board (SGI, Sun,
> anything)
> > > and pretty much all the software (Chap-C, etc.)
> > > * Sritek VersaCard - missing the MC68000 Xenix and PC interface
> software.
> > > * Sun FDDI/DX (VME) - missing the SunOS driver tape
> > > * Sun GT - busted, missing much in the way of hope tracking down the
> > > fault, never mind repairing
> > > * Sun TAAC-1 - missing the software
> > >
> > > A goodly measure of IBM RT AOS/4.3 software has been recovered and
> > > archived in the last couple years. Some of it from my own efforts.
> There
> > > are enough of the 6152-specific pieces that exist in situ to make a
> usable
> > > 6152 system, but they're not complete. It???d be nice to turn up some
> of the
> > > official distribution media for that. There???d have been a QIC tape
> adding
> > > the 6152-specific kernel pieces, maybe a floppy or two with the DOS
> and/or
> > > OS/2 host components (if they weren???t also on the tape).
> > >
> > >
> > > ok
> > > bear.
> > >
>
> --
> ---
> Larry McVoy Retired to fishing
> http://www.mcvoy.com/lm/boat
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.tuhs.org/pipermail/tuhs/attachments/20250719/45ba15dc/attachment.htm>
More information about the TUHS
mailing list