QoS priority - 'real-time' vs 'high'

[psycoma1984]

Hi,

What is the difference between 'real-time' and 'high' priority than? Will there be any difference under similar circumstances when you have 100 Mbit/s MAX egress and 50 Mbit/s guarantee, while actual traffic matching this class exceeding 100 Mbit/s? Previously I was under impression that real-time one will just drop traffic exceeding queue limit and anything else will just reclassify everything above to default class 4 as bandwith is available there. 

But after checking https://live.paloaltonetworks.com/t5/general-topics/qos-is-quot-real-time-quot-priority-a-strict-pri... and doing my test it seems like no reclassification is happening in both cases - doesn't matter which priority you use (except maybe low) - once all bandwith is used for a traffic within this class, it will just drop assigned kind of traffic.

Maybe @michaelmertens can share results of his QoS implementation?

What is a correct assumption?

Thanks

[reaper]

the real-time classes have a dedicated hardware queue that bypasses the regular queue when resources are depleted (and packets need to fight for resources), all the other priorities are 1 shared resource pool with higher priority packets getting accepted and forwarded first and low priority packets getting discarded first/forwarded last.

Priorities only come into play when resources are depleted/interface congested, not when limits are being enforced

[psycoma1984]

Hello,

That's what I am alluding to really. If link is congested - what happens to a traffic in a class with a real-time priority, where volume of traffic egressing interface for this class higher than egress value mentioned in a profile for this class? Can PaloAlto reclassify it to best-effort class 4 or will just drop it?

Also another question - what happens if we have more than 1 real-time class in a profile? 

PS: Actually Steve Cantwell in this thread says something different about whether queues are been enforced same time as qos profile applied: https://live.paloaltonetworks.com/t5/general-topics/qos-is-quot-real-time-quot-priority-a-strict-pri...

Q: Does QoS queuing ONLY get engage when the interface is congested? 

A: the answer is NO

Thanks. 

[reaper]

Real-time queueing has priority over all other priorities, lower priorities will always be dropped before real-time

you can put more classes into real-time priority, but putting too much of your mean traffic in this priority will simply create a FiFo queue, defeating the purpose of real-time over lower priorities

The queue would indeed also be used if you run into a limitation for the class, but it would be a bit odd to have a limit _and_ real-time queueing in the same class i think?