[Next] [Previous] [Up] [Top] [Copyright] [Books About Ethernet]

1.5 Operation of Ethernet

Collisions


If more than one station happens to transmit on the Ethernet channel at the same moment, then the signals are said to collide. The stations are notified of this event, and instantly reschedule their transmission using a specially designed backoff algorithm. As part of this algorithm the stations involved each choose a random time interval to schedule the retransmission of the frame, which keeps the stations from making transmission attempts in lock step.

It's unfortunate that the original Ethernet design used the word "collision" for this aspect of the Ethernet medium access control mechanism. If it had been called something else, such as "stochastic arbitration event (SAE)," then no one would worry about the occurrence of SAEs on an Ethernet. However, "collision" sounds like something bad has happened, leading many people to think that collisions are an indication of network failure.

The truth of the matter is that collisions are absolutely normal and expected events on an Ethernet, and simply indicate that the CSMA/CD protocol is functioning as designed. As more computers are added to a given Ethernet, and as the traffic level increases, more collisions will occur as part of the normal operation of an Ethernet.

The design of the system ensures that the majority of collisions on an Ethernet that is not overloaded will be resolved in microseconds, or millionths of a second. A normal collision does not result in lost data. In the event of a collision the Ethernet interface backs off (waits) for some number of microseconds, and then automatically retransmits the data.

On a network with heavy traffic loads it may happen that there are multiple collisions for a given frame transmission attempt. This is also normal behavior. If repeated collisions occur for a given transmission attempt, then the stations involved begin expanding the set of potential backoff times from which they chose their random retransmission time.

Repeated collisions for a given packet transmission attempt indicate a busy network. The expanding backoff process, formally known as "truncated binary exponential backoff," is a clever feature of the Ethernet MAC that provides an automatic method for stations to adjust to traffic conditions on the network. Only after 16 consecutive collisions for a given transmission attempt will the interface finally discard the Ethernet packet. This can happen only if the Ethernet channel is overloaded for a fairly long period of time, or is broken in some way.


Quick Reference Guide to the Ethernet System - 04 SEP 95
[Next] [Previous] [Up] [Top] [Copyright] [Books About Ethernet]

Generated with CERN WebMaker