Peter, thanks for your new posting!
Again, yea, it looks like we have different view about what is, or is not, a collision domain. I believe I understand what you're stating, but again, how can there be a bound on something that cannot happen at all? For example, we both agree two "shared wire" network segments, separated by, for example, a bridge, are indeed bounded by the bridge. Yet, change those two network segments, into segments where collisions are impossible, what's there to bound? Or, in other words, when you write ". . . I tend to treat a "domain" as a "bounded scope of impact", . . .", yet, again, if there is no possible impact, means, to me, you cannot define the scope.
Of course, laugh, perhaps you take more of a Calculus approach. I know when introduced to Calculus, I wondered, then, and now, how you can compute an actual area of that's bounded by infinity. I.e. the area is never "closed", because it's infinite, but when you get there, the area is exactly X. Then there were also equations which you could solve for the surface area of a solid, but not its volume. Oh, my, that still causes my head to hurt.
In this case, I think we'll agree to disagree on that, but not a disagreement that's of any concern, at least, to me.
Okay, regarding you providing a possible impression of overlooking, importance and/or implications, yes, I, at least got that impression. (Oh, and as an aside, not any impression of lack of knowledge or understanding.)
What I've been trying to highlight is, shall we say, the "quirks" of Ethernet designed for connecting to hubs, vs. connecting same to switches.
You, yourself, in your last post explicitly mention 3 hosts connected to a hub, with two transmitting, cause an issue. (You also correctly describe that physically, since a hub is a digital repeater, the "signals" on the wire would not be an analog mess as would be found on a real shared wire, like 10Base-2 or -5.)
What you didn't mention is two hosts would work just fine, transmitting concurrently, or could, but they won't! This because of the self feedback designed for the 3 or more hosts case, causes the 2 host case to fail too.
The implications of this, on a switch, which is physically 2 hosts/NICs, it doesn't need to fail, but it does, to continue to work as it does, correctly, for 3 or more hosts on a hub. This design causes "collisions" with switches, in HD mode, although they are not physically possible.
FD mode, basically, "informs" the connecting (2) hosts/NICs, they no longer need to self loopback avoiding "collisions" that are, already, not physically present.
It's this point, of, shall we also say, of "backward compatibility" that you appear to be overlooking (again, just to me, and further, something, I suspect you fully understand).
Conversely, when discussing HD and FD, you touch upon mixed (HD and FD) usage between two NICs, and or how switches would not forward corrupted frames (although do I recall [?] some cut-through switches can, to some degree do that).
Unfortunately, the written word, sometimes, does not exactly convey what the author intended. (I know my poor prose often causes that.)
So again, to your statement "I am not aware that I have overlooked that, or that I gave off such an impression.", also again, to me you did, but such could be wholly my fault. Conversely, what I've been trying convey, and (again, my impression) failing at, likewise, could totally be my fault.
In other words, other than our view of how to define a CD, we may be in 100% agreement.
As another aside . . .
I'm old enough to have lived through some major network technology changes, and their impact in large production environments.
For example, a couple of decades ago, I was working in a Fortune 100 company that was Big Blue, and for networking, still using token-ring. Which, more-or-less they were happy with, but when TR NICs were still as expensive as current generation PCs, the latter including built-in 10/100 Ethernet NICs, they finally decided to upgrade from 4 Mbps TR to "faster" 10Base-T (on hubs).
I wasn't part of the Network group, but I convinced my group's management to "defer" upgrading their TR to Ethernet, which they accomplished.
Well, you should have heard the dismay, from other parts of the Enterprise, regarding the "lousy" network, at least on those parts "upgraded" to 10 Mbps Ethernet.
After enough of that, they upgraded to 10 Mbps switches. I then recommended moving to it; which also removed all the performance complaints (except when they did have a duplex mismatched host/switch combination). (BTW, for most user hosts, on a switch, performance didn't really matter, hugely, whether host/switch was HD/HD or FD/FD, as user hosts didn't generally have huge amounts of concurrent bidirectional data transfers.
Reason I mention this, I'm seen the actual impact, or not, of collision domains and HD and FD, hubs and switches.
