Mritunjay, Leo,
Please allow me to join.
With respect to collisions on a switch, and Mritunjay's original question whether collisions occur in switches, I think it's worth to revisit the context of collisions themselves - since what we call "collisions" on switches are not the full-bloodied collisions we used to know on hub and coax ;)
The term "collision" in Ethernet was coined along with its absolute beginnings when Ethernet was running over a shared medium (the respectable 10Base5 "thicknet", and later 10Base2 "thinnet", both creating a huge bus with all hosts attached to it). A collision occurred when two or more stations have been transmitting data
- at the same time
- over the same shared medium
The obvious reason why this was a problem was that the senders' signals overlapped, and got "garbled" - listening stations could not understand the resulting "mix".
Later, with 10BaseT, the cabling was changed to twister pair, and suddenly the medium itself stopped being shared: There were separate pairs of wires for transmitting and receiving signals, so the medium itself was not creating collisions anymore, as opposed to a coax wiring with all hosts connected to it. With hubs, however, the function of the bus still remained within the hub - if, say, stations A and B were transmitting at the same time, the hub would receive and repeat all received signals as they came in, so a station C connected to the same hub would again hear a "mix" of all signals, again meeting the notion of what a collision is.
With switches, however, this kind of electrical signal overlap is not possible anymore. The medium itself - either fiber or TP cabling - is not the source of collisions anymore since it is not shared, and a switch is capable of storing the incoming frame into a buffer if the outgoing port is currently busy transmitting another frame, so a switch also does not "mix" the incoming signals when sending them out. In terms of signals, with switches, true collisions as we knew them no longer exist.
However, what Leo described absolutely correctly is what we understand as "collisions" nowadays with switches: A talker overlap resulting of a misunderstanding whether it is okay to speak when spoken to (that is, a duplex mismatch). If the device on one end of a link expects that the link is silent during its own transmission (half duplex) while the other device does not care and sends data whenever it wants (full duplex), the half duplex device will consider and report these talker overlaps as collisions. What is different here is that the signals themselves were never "garbled", they never ended up in an unintelligible mix - just the half duplex device considered it inappropriate to be interrupted when sending data itself, and that's why it both dropped its own transmission, and whatever was received from the other side during this occurrence.
So, yes, with switches, we do have collisions, and they do result in frame discards, but not due to the frames being unintelligible by multiple signals overlapping each other on the same medium, but rather, due to violating the perceived half-duplex agreement.
Best regards,
Peter
