4

I'm just revising for an exam on Networks and Data Communications, and there's one thing I don't get about CSMA/CD and Ethernet. It's supposed to be fairly stable, for instance if a computer drops out of the network, it's not a problem like it might be in a token ring network (I think).

But Ethernet works by all the other computers waiting for the currently transmitting computer to finish what it's doing, and then the others use CMSA/CD to determine who goes next. What if one computer malfunctioned and kept sending a continuous stream of data in an infinite loop? In fact, is there a standard time for pcs to transmit before they yield to others?

3 Answers 3

14

If there is something generating noise on the network continuously, the network will be broken for all systems on that network.

There are a couple of conditions that must be met:

  1. If you are on a switched network, the "noise" must be in the form of an Ethernet packet. If it isn't a valid Ethernet packet, then the switch port receiving the noise will not forward it to any other port.
  2. If you're on a truly shared media, the noise may be anything such as a 10b2 cable strung over a broken fluorescent light fixture.
  3. If you're on a switched network, the packets must be addressed to either an address that isn't in use or to a broadcast address. The switch will "flood" the unknown address to all ports and only stop flooding if it gets a response.

A common example of this sort of failure mode is with broadcast storms.

2
  • +1 Very concise, great examples
    – Dave M
    Jan 21, 2010 at 15:52
  • Broadcast storms and jabber. Does a switch transmit this back into the network? I've had instances in a multi-segmented network where everything on the switch was knocked out by one machine, but it didn't make it back through the uplink port to affect anything else on the other switches. Jun 11, 2012 at 15:28
5

One should also mention that CSMA/CD is not required in an age of where everything is switched. The switch makes the medium look like it isn't shared since it implements a point to point topology.

2
  • 1
    Exactly, now that nobody uses hubs, testing on this kind of thing mainly just serves to provide some historical perspective on, say, WHY nobody uses hubs anymore. ;) Jan 21, 2010 at 18:40
  • 1
    I certainly don't miss the "who unplugged the terminator" days! My answer isn't really 100% correct since if you've got some crazy system pumping out billions of tiny packets at wire-speed, some clients may see some small fraction of available bandwidth, but they'll still think the network's "broken". The effect is much worse with network loops and broadcast packets because you'll have more than one device generating the traffic. Failure modes can get complex fast, but my answer is 90% to 95% correct... Not correct enough to merit so many upvotes, but ...
    – chris
    Jan 21, 2010 at 19:18
1

Really a comment on @chris's answer (which I upvoted): That applies equally well to non-Ethernet networks, though they have different failure modes. Continually issuing or destroying tokens, for example, or not forwarding packets properly.

You must log in to answer this question.