QOS role Query

[knaik99]

Hi all,

Just want to know QOS come into action when there is 100% link utilization or link congestion.

otherwise traffic can flow freely from link right?

20% traffic is assigned for Voice but Voice can travel freely if link is not overutilized right?

[MHM Cisco World]

The core idea of QoS is that it will prefer some traffic to forward than other.

[Joseph W. Doherty]

Hmm, not exactly.

What QoS core idea is what those letters mean, i.e. Quality of Service.

Certainly we very often do prioritize forwarding some traffic, but that's just one tool or technique.

Or you might say the core idea of QoS is to remove the "maybes" in my answer to OP's questions.

[Joseph W. Doherty]

It depends on the kind of QoS.

For example, policers and shapers, generally, engage before 100 percent.

Priority kinds of QoS engage when traffic is queued, which doesn't require 100 percent link utilization nor at 100 percent link utilization (BTW, link utilization is often misunderstood).

Technically, the answer to your questions is: maybe, maybe not.

[Joseph W. Doherty]

BTW, my prior answers to your questions, although I believe correct, are likely to leave you with additional questions.  If so, feel free to query further.

As I see you have a Cisco Professional level certification, you've no doubt have been exposed to some level of QoS, especially in supporting real-time traffic like VoIP, but much QoS education material, IMO, does a very poor job of truly making (full) sense of QoS.

Such poor QoS educational materials often lead to questions, like yours, which at a professional certification level, possibly, you should already know.  (Note - this last is not intended as a negative to you [ditto @MHM Cisco World ], just again, IMO, most QoS educational materials, I believe, are poor.)

[husseino]

Correct, you assign traffic to different queues, you allocate BW % per queue. When there is no congestion, traffic can go beyond allocated BW to the shaped value of the interface.

[Joseph W. Doherty]

"Correct . . ."

Again, a correct answer to OP is very much "a depends" kind of answer, well at least "real world".  Unless questions are just in the context of something like a certification exam question, then the following might not need to concern us.

Otherwise, OP's questions are too generic to provide a "correct" answer without getting into much deeper details about actual QoS policy applied, platform it's applied on, actual traffic behavior and even what exactly is meant by terms like "100% link utilization", "link congestion", "traffic can flow freely" and "link is not overutilized".

As an example if given the following CBWFQ policy:

policy-map example
class Mission-critical-traffic
bandwidth percent 40
class class-default
bandwidth percent 20

And asked: What's the guaranteed minimum bandwidth percentage for class "Mission-critical-traffic"?

And answer choices are:

a) 0%
b) 20%
c) 40%
d) 60%
e) 100%
f) none of the above

What's the correct answer?

[husseino]

The question answers are incomplete (should say % of what), answer is 40% of the interface shaped BW.

[Joseph W. Doherty]

(Firstly, @husseino, thank you for responding.)

"The question answers are incomplete (should say % of what) . . ."

% of what?

Whatever the provided policy is applied to.  (Percentage based bandwidth settings was, as you might know, one of the earlier enhancements made to CBWFQ [originally, bandwidth setting were limited to using a bps setting], so that the same policy could be applied to interfaces of different bandwidths.)

The shown policy uses bandwidth percentages, and question asks what's the bandwidth guarantee, for one of the policy's classes, using a percentage answer.

As you further note the answer is 40% (answer "c"), that's exactly the kind of answer I was looking for.  Again, being percentage based, "interface shaped BW" is extraneous/superfluous.

Does question make sense now?  Do you agree "interface shaped BW" is extraneous/superfluous?

Anyway, again, you did select a percentage answer, the 40% matching the bandwidth percentage assigned to the class that was asked about.

However, I believe 40%, although very much appearing to be the correct (and/or the expected) answer is actually incorrect!

Further, if this was a question exactly like this on a certification exam, I wouldn't be surprised that such an exam would also consider 40% the correct answer, although, again, I believe it's incorrect.

Assuming you want to stand by your belief that 40% is correct, while I stand by my belief it's incorrect, hopefully one of us will learn why the other answer is correct.

Would you like to explain why you believe 40% is correct, or would you prefer me to explain why I believe it's incorrect?

[husseino]

I would like to know the right answer

[Joseph W. Doherty]

I believe the correct answer to the question I posed, for the policy presented, would be "f) none of the above".

I believe the correct percentage is 2/3, or about 66.7%, not 40%.

Why?

Because only two classes were defined (one being class-default, always present, whether explicitly defined or not), with 40% defined for "class Mission-critical-traffic" and 20% for "class class-default".

Only 60% was defined, 40% was not.  So?

Between those two classes, they use their two bandwidth allocations, in this case, as percentages, to allocate dequeuing weights (the W in CBWFQ) in the ratio of 40:20 or 4:2 or 2:1.  I.e. "class Mission-critical-traffic"'s traffic bandwidth/volume will be dequeued twice as much as "class class-default".  As percentages it's about 66.7% and 33.3 % for those two classes.

If we revise the policy to be:

policy-map example2
class extra
bandwidth percent 40
class Mission-critical-traffic
bandwidth percent 40
class class-default
bandwidth percent 20

Now "class Mission-critical-traffic" does have a minimum guarantee of 40% because all 100% is defined.

Yet, even in this case, the forgoing (the minimum guarantees) is only seen if all 3 of those classes want their bandwidth percentage or more.  If the newly added "extra" wanted zero bandwidth, and "class Mission-critical-traffic" and "class class-default" wanted all the possible bandwidth, again, those two classes would divide the available bandwidth 40:20, 2:1 or 2/3:1/3.

So, a very big difference in what the original policy and the revised policy "guarantee" to their classes although at first glance they look they would guarantee the same results.

Of course, since the first policy I posted actually guarantees 2/3 and 1/3 bandwidths, you could argue that any lessor percent is also guaranteed, i.e. 66.7% insures 40% and 33.3% insures 20%, but is that what comes to mind when you see a statement like "bandwidth percent 40"?

Or, perhaps more problematic, you do add the "extra" class, which does indeed want its 40%.  Possibly then you have problems with either or both of the two prior classes now having performance issues, yet their bandwidth guarantees haven't changed(?).

Ah, but in fact, they have actually changed.  Consider if you originally had allocated those two classes as 67% and 33%.  To add a third class with 40% means you need to reduce the bandwidth guarantees of one or both classes.  It's now obvious you're reducing actual guaranteed bandwidths.

BTW, if you're wondering why I keep noting "I believe", it's because Cisco, in their later IOS versions may have revised how CBWFQ does dequeue (and if so, I'm unaware of such revisions), but the forgoing is how CBWFQ has worked for years (and WFQ).

I noted in an earlier posting, in this thread, I consider much (even possibly most) QoS educational material "poor".  Basically, my prior posting's policy was a "trap" in the sense that much educational material will teach about setting a bandwidth floor guarantee, but not much touch on the subjects of what if you don't actually explicitly allocate all 100%, or what happens when there's excess bandwidth to use because not every class is using all it can at any one one point in time.

This is only a small example of how, IMO, so much is often not emphasized, or even mentioned, in QoS training materials.

If you believe I'm mistaken, or have additional QoS questions, please post them.