From my understanding, full duplexed switched ethernet networks provide determinism. However,

the Bandwidth Allocation Gap does not define a time schedule 2 virtual links could have frames transmitting at the same time and 1 virtual link will have to wait

does jitter in the network mean that it cannot possibly be deterministic due to the possible contention?

  • Did any answer help you? If so, you should accept the answer so that the question doesn't keep popping up forever, looking for an answer. Alternatively, you can post and accept your own answer.
    – Ron Maupin
    Dec 17 '20 at 16:19

When congestion happens at a bottleneck in a switched network, bandwidth is (somewhat) evenly shared by default. For instance if you interconnect two switches with a single 1G link and then try to transmit two flows of potentially 1 Gbit/s across those switches, each flow will get 500 Mbit/s pretty exactly.

With most switches, congestion causes frames to be dropped by source port: one server trying to send one flow of 800 Mbit/s and another server trying to send two flows of 500 Mbit/s across the two switches from above will end up the one stream at 500 Mbit/s and the others at 250 Mbit/s each - half of the total bandwidth for each source port. However, this depends on the exact hardware at hand and should be tested before relying on it. The effective throughput also depends heavily on any congestion control that takes place on higher network layers like in TCP.

If you need more control, you can use QoS settings to prefer one kind of flow over the other kinds (or define different priority classes). How this exactly works also depends on the hardware you're using. Some switches support more priority classes than others and some support more scheduling schemes than others.

Jitter also depends extremely on the exact pieces of hardware.

  • Thank you for the response. The architecture I am studying, AFDX, uses QoS control with traffic shaping function. Virtual channels are implemented as uni-directional source to destination connections that each have a pre-defined smallest and largest ethernet frame size. There is also a bandwidth gap defined in software for each virtual channel that sets a delay between sending frames. The bandwidth is guaranteed for each channel; however, multiple virtual channels can share one ethernet link. Does this make real-time control impossible; thus, making this a non-deterministic network??
    – user40590
    Mar 18 '20 at 17:00
  • "Real-time" is relative. You'll need to compare your expectations with the hardware specs.
    – Zac67
    Mar 18 '20 at 18:11

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.