Collisions domain

[gattina59]

Hello everyone on a question ccna I found different answers and I noticed there is too much confusion seems that the answer is subjective ……

If I have two switches connected in the same broadcast domain then switch A link switch B we have 3 hosts connected.

There is obsiously a link between switch A and Switch B.

The answers are different there are those who say 7 and there are those who say 6. Who says 6 also gives the motivation; He says why the switch to switch connection is full duplex and when it is full duplex is not considered.

The following statement is written in the ccna book: In a modern Lan with all switches and routers, even thought full duplex removes collisions, think of each link to a separate collision domain when the need to troubleshoot arises.

Who should we believe?

[Joseph W. Doherty]

"Either way, we know that there is NOT 1 collision domain :-)"

Sorry, incorrect.  However, I wonder if such a misunderstanding is why the original question had so many collision domains.

Why?

Ok, you're correct about there being one collision domain on shared media Ethernet (10Base2 or 10Bast5).

Before going to 10BaseT, what's an actual collision on shared media?

It's when while one NIC is transmitting, another NIC (or NICs) begins to transmit.  When the two (or more) signals are on the same wire at the same time, the "collision" of the multiple transmissions corrupts all, effectively no other NICs can decode the transmission.  (BTW, of course, CSMA/CD Ethernet NICs first "listens" for a clear media before transmitting, but that only avoids collisions if an active transmission has yet reached the NIC.  Getting into the "weeds", it takes time for a signal to be "heard" by all attached NICs on the same wire.  BTW, these timing requirements also come into play with why there's limitations on overall length of a common shared media [including why there's a limitation on how many 10BaseT repeaters you can use - also why it less for 100 Mbps].)

On 10BaseT, you can only have two NICs on the same media, as one NIC's TX connects to the other NIC's RX.  This being the case it's physically impossible to have a physical collision, as described above.  However, when moving to 10BaseT, and to keep to original shared media design, NIC's logically emulate physically collisions.  I.e. When a NIC is sending on its TX, if it then also receives on its RX, then its "logical" collision is treated as a "physical" collision.  So, the behavior of two NICs, on 10BaseT is like two NICs on 10Base2 or 10Base5. I.e. the two physical wires create but one collision domain.  Not zero and not two domain, again one because it's logical.  To further clarify, if both full duplex pair NICs are transmitting, each will receive the other's transmission, and since both are transmitting, both will act as if a collision has happened.

When you connect multiple hosts to a hub, the hub repeats one port's transmission to all other actively connected ports.  So, again, although there's no physical collisions anywhere, the hub (or hubs), and their hosts, act like they are all on the same physical shared wire, so there's there's still only one collision domain.

Replace the hub with a switch (still following the half duplex "rules"), but now you don't need to immediately regenerate/transmit a received frame from the switch.  You can queue/buffer them, and send them when you can.  Instead of having one (logical) collision domain shared by all hosts, each switch port is its own (logical) collision domain.  I.e. the switch port and it's connected host can (logically) still have a collision, but there will not be collisions between hosts.

The next step, someone clever, looking at 10BaseT, thought, as there are no physically collisions, why emulate them?  We don't for other full duplex Ethernet (i.e. fiber).  So, with full duplex, there is no longer a logical collision.

On a switch, which support both half duplex and full duplex, if you had three hosts, any host running in half duplex would be in its own (logical) collision domain with the switch port.  Any host running full duplex is not in any collision domain.

[Teacher Todd Booth]

@joseph>  Having a powered off switch sort of creates a "trick" question.

At least for 99% of my students who take the first course in networking, it is clearly and obviously not a trick question. If the switch is powered off, there are no active collision domains.

The same goes for full duplex with dedicated wires in each direction. 

At least for 99% of my students who take the first course in networking, it is clearly and obviously not a trick question. Under the above assumptions, there are no active collision domains.

@joseph> By design, there's still two collision domains.

Your comment is very vague and ambiguous, which is why we can't come to an agreement.  For example:

By design, when running if full duplex with dedicated wires, there are no collision domains.

By design, when running if half duplex with shared wires, there are multiple collision domains.

So by you being vague and ambiguous, you are simply not addressing that, based on the specific assumptions, there are multiple valid answers.  For example, with 2 different assumptions, we can have two entirely different valid answers.

> However, what can cause confusion if the environment isn't fully defined. Perhaps that's the point you're trying to make?

Yes, that is exactly my point.

1. If environment (based on assumptions) A, B, C, the only valid answer is X.

2. If environment (based on assumptions) D, E, F, the only valid answer is Y.

The above two questions and answers do not conflict (there is not one common correct answer), but the people posting in this topic don't seem to understand the above points 1-2.

Thanks for your helpful comments!!!

[Joseph W. Doherty]

"At least for 99% of my students who take the first course in networking, it is clearly and obviously not a trick question. If the switch is powered off, there are no active collision domains."

Ah, but now you've added a very important word "active" as in "no active collision domains".

Generally, in my experience, we don't discuss network design when devices are powered off (excluding perhaps what happens when a network device does go off or on-line). So, when you ask how many collision domains are there for two hosts connected to a powered off switch, most practicing network engineers, I believe, would assume (BTW, ever hear how the word "assume" might be parsed?) you're concerned with the design topology. Otherwise, with the switch powered off, what's to discuss, operationally, beyond perhaps how to power the switch back on?

That's why I consider such a question a "trick question", even though you're operationally correct, it's not generally, I believe, what's considered in real world networking.

Of course, I too wouldn't expect first time networking students to consider this a "trick question", as what do they yet know. ;)

"@joseph> By design, there's still two collision domains.

Your comment is very vague and ambiguous, which is why we can't come to an agreement."

Laugh, well yes, taking that sentence, as often the case with any individual sentence, out of context, will often not convey the correct/full meaning.

"For example:

By design, when running if full duplex with dedicated wires, there are no collision domains.

By design, when running if half duplex with shared wires, there are multiple collision domains.

So by you being vague and ambiguous, you are simply not addressing that, based on the specific assumptions, there are multiple valid answers. For example, with 2 different assumptions, we can have two entirely different valid answers."

Still vague and ambiguous, eh? Ok, please provide design examples, in the context, as earlier described, of a switch, with two or more hosts, each on a separate switch port, that those two statements are not true and/or how they are vague and ambiguous.