In Cisco packet-tracer while simulating it is shown that switches receive a whole packet before starting to transmit it to the next switch. I guess real switches can perform better if they forward frames as they receive them and not wait for all frames of a packet to come. Is the behaviour shown in packet-tracer accurate or switches forward frames as soon as soon as they get them?

  • 1
    Note that switches handle frames, not packets.
    – JFL
    May 28 '19 at 19:12

The standard mode for switch operation, as defined in the 802.1.D standard is store and forward, in which, as you described, wait for the full frame to be received before forwarding it.

This is still the way most switches work.

As you guessed, it impose some delay, thus increasing latency.

That's why high-end switches use cut-throughmode. In this mode a switch can begin to forward the frame as soon as it received the destination mac address (which is the first field received after the preamble).

Although it is a violation of the standard, this violation is not dramatic, and provide some gain in latency. However with current wire speeds, for most applications this gain is not significant, in the order of a few microseconds.

Also, there's some drawbacks, since you cannot perform deep inspection of frame or the packet contained in the frame, and you will not detect checksum error before transmitting a frame (the receiving end will discard the frame when computing the checksum but you wasted some bandwidth).

  • 2
    provide a real gain in latency - I would challenge this statement in the general sense. Only in the time it takes to transfer the frame payload (i.e. excluding the header). At today's wire speeds, unless you are using jumbo frames, the gain is almost negligible.
    – YLearn
    May 28 '19 at 12:30
  • 1
    @YLearn Seconding that - with Gigabit Ethernet, a frame has a maximum serialization latency of 1518*8 ns = 12.1 µs. 10GE reduces that to 1.2 µs and so on.
    – Zac67
    May 28 '19 at 17:41
  • @YLearn and Zac67 thanks, I have edited.
    – JFL
    May 28 '19 at 17:55
  • 1
    Tell it to the high speed trading guys. (ns's matter)
    – Ricky
    May 28 '19 at 20:40
  • JFL Thank you very much for your answer and also referencing the standard. @YLearn I think with cut-through switching the network depth(number of switches in between two nodes as I know it) will make minimal impact but with store and forward we must be conscious of this depth. Am I right?
    – S.Rad
    May 28 '19 at 21:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.