Ethernet Collisions

Neeko

Odom states that "collisions occur as a normal part of the half-duplex logic imposed by CSMA/CD, so a switch interface with an increasing collisions counter may not even have a problem. However, if a LAN design follows cabling guidelines, all collisions should occur by the end of the 64th byte of any frame. When a switch has already sent 64 bytes of a frame, and the switch receives a frame on that same interface, the switch senses a collision."

Can someone explain this please? I understand half/full duplex logic, micro segmentation and all that, but not sure why 64 bytes is a significant figure in this case.

ccnacertified2000

Let me explain this.

Ethernet started out with a standard based on coax cables sharing the same medium. That standard has a max end to end cable length of 500 meters. To make sure collisions are heard by all of the hosts including the sending host, the sending host needs to know a collision happened before it sends the last bit otherwise, it won't know its own frames collided. Because of the speed that it takes to send a bit from one end to another, there is a minimum amount of time a host needs to be sending in order to detect its own collision. This ensures that by the time it sends the last bit, the first bit will be heard by everyone in the same medium already. This minimum of time is about 50 microseconds i believe. At ethernet speeds, this translates into the 64Bytes of data. So assuming the host is sending at 10Mbps, and is required to send for 50 microseconds, the amount of bits sent in 50 microseconds in 10Mbps speed would translate to about 64 Bytes. Theres a lot of simple algebraic math involved to calculate the 64 bytes. You need to know the speed of electricity through the coax cable, the ethernet timing for sending bits, and the length of the cable (500m), and using all that you can calculate that the minimum would be 64Bytes.

But we have 100Mbps UTP cable now, and gigabit too, why is it still 64 Bytes? Compatibility. We have switching too... Compatibility to enforce the collision even if none happened to maintain compatibility with regular ethernet networks.
This is also why you can only have a maximum of 4 hubs/switches chained, creating 500M max end to end length of ethernet. There is also a maximum amount of time a hub can delay a packet to maintain this standard.

Its all electrical engineering. If you study it closely and calculate, you will find it truly amazing how it works out to 64 Bytes and why it makes total sense.

Neeko

So according to LAN cabling guidelines as Odom puts it, the recipient(s) of the frame should receive the first bit and send a jamming signal if a collision has occurred, all within the time it has taken the sender to send 64 bytes?

So basically if you get a load of late collisions on a switch interface there is some crazy half duplex **** going on or the LAN is too long?

Actually, I've just realised. If the whole LAN was too long that would indicate the whole thing is a shared medium. A better way to put it would be to say the single cable connected to the interface in question or the collision domain is too long.

Good old Wendell, assuming I know the ins and outs of everything lol.