QoS- a (probably) basic question

Hey folks!

This is probably a pretty basic question for you all, but hey... we only learn by asking eh.

My primary background is on AV networks where we are dealing with equipment that requires precise timing & reliable delivery- so QoS is almost a bit of a buzzword in the industry. In general we are dealing with PTPv1/2 clocking traffic tagged as 46/EF and sent to the highest priority queue, and then some flavour of media- usually tagged as 34/AF41 and sent to the next queue.

Now, my question is this. In general for VoIP etc, whilst studying I’ve noticed the emphasis is generally on QoS only really being a concern when bandwidth becomes tight and decisions have to be made about what gets sent, queued or dropped based on priority- makes sense especially if you are going over a WAN connection- which is not relevant to me. My systems individually all live on a single L2 domain, and based on the normal manufacturer recommendations, on the switches I run strict priority queues- but I’m curious about when the traffic is only traversing across the backplane of a single switch. At this point, bandwidth isn’t a concern- we certainly don’t get close to needing gigabit links on any individual device. That being said, when trunking between switches we typically only use a 1GB link as it’s all that’s needed for the traffic- but we are obviously getting closer to bandwidth limits at that point.

So... I did some experimenting. Normally, I have the ACL to identify the specific traffic and send to the correct queue applied at the ingress port of every access & trunk port on the switch. Obviously this works just as expected. Just for fun, I instead put the traffic into the correct queues on the egress & ingress of every trunk port- so any traffic that isn’t leaving that particular switch and is just heading to another port doesn’t get any special treatment. Getting any meaningful information from the hardware isn’t too easy, however everything still functions as expected, and I saw no difference in PTP clock offsets between the ‘normal’ setup and my experiment.

Now, my gut tells me that the original setup is the best practise; and that if traffic needs to be given any special treatment it should happen from ingress to the switch- but is that correct?

Is it reasonable to say that, within a single switch, QoS is ‘less important’ because, in most cases, bandwidth across the switch will be much higher than between switches?