19 Jul
2025
19 Jul
'25
7:11 p.m.
On Sat, Jul 19, 2025 at 11:53:20AM -0700, Tom Lyon wrote:
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.
Huh, that's the first I've heard of that. My personal experience was
that the issue was signaling at 100Mbit had cross talk issues. Until
it didn't.
And then there was HP with 100Base-VG (iirc). Sigh.
You and Andy understood that faster Ethernet was all that was needed.
What I wanted was ethernet packets. I'd seen switch technology and
realized that you don't really need to store and forward (unless the
outgoing port is busy) so people were starting to talk about sending
the first bits out the outgoing port before the last bits came in the
incoming port. The nay sayers were mumbling about forwarding corrupt
packets but that got shut down because (A) the final destination of
the packet will catch that it is corrupt and (B) corrupt packets are
vanishingly rare so making all the switches slow for something that
doesn't happen often is stupid.
The other thing I realized is that ethernet packets are sized just fine.
I used to want jumbo packets (8K + space for headers) so you could do
SGI's page flip / copy on write. But infinitely large packets means
infinitely large buffers, SGI's numa interconnect educated me on that.