QoS - policy-map interface interpretation

[AlejandroJuarez03280]

Dear community

Can anyone help me to understand the QoS command "show policy-map interface in ios XE? 

In each queue i can see 7 classes and justo some of them have packets, do you know how is the traffic balance in this classes or how to interprete this output?

Below an example of the output and the configuration, It is just testing in a lab to understand QoS:

class-map match-any BUSINESS
match qos-group 5
class-map match-any CD
match qos-group 1
class-map match-any MGMT
match qos-group 6
class-map match-any Default
match qos-group 2
class-map match-any VIDEO
match qos-group 4
class-map match-any Queue1
match qos-group 1
class-map match-any Queue0
match qos-group 0
class-map match-any Queue2
match qos-group 2
class-map match-any Queue5
match qos-group 5
class-map match-any Queue4
match qos-group 4
class-map match-any Queue6
match qos-group 6
!
policy-map Escobedo_Classes
class Queue0
priority level 1
police rate percent 9
class Queue1
bandwidth remaining ratio 30
random-detect
class Queue4
bandwidth remaining ratio 15
random-detect
class Queue5
bandwidth remaining ratio 13
random-detect
class Queue6
bandwidth remaining ratio 31
random-detect
class class-default
bandwidth remaining ratio 2
random-detect

I appreciate for your help, thank you

Regards

[Joseph W. Doherty · Accepted solution]

"In each queue i can see 7 classes . . ."

Actually there are 8 classes, 0..7.

". . . and justo some of them have packets, do you know how is the traffic balance in this classes or how to interprete this output?"

I do know how to interpret, but not knowing your level of QoS knowledge, unsure how good my explanation will be for you.  (Feel free to ask follow up questions.)

Firstly, the classes you are looking at, in each class queue, are IPPrec (first 3 bits of ToS byte) based WRED classes.  Basically, each of those classes examines the current overall average queue depth, examines the packet's IPPrec marking, determine for a particular marking whether the average queue depth is less than minimum drop threshold, within the minimum to maximum RED drop thresholds, or over the maximum RED drop threshold.  For the less than minimum drop threshold, packet will not be dropped.  For over the maximum threshold, packet will always be dropped.  When between minimum and maximum, a chance percentage of drop is computed, and if the packet is a "winner" (or perhaps better described as a "loser"), packet is dropped.

Since this is WRED, each RED class can have different min and max thresholds, and a different max threshold drop chance percentage.

As to "balance", each or your RED classes have the same max and max drop percentage, but low valued IPPrec values have a lower min drop threshold.  I.e. they are more likely to be have their packets dropped at a higher rate.  I could compute a relative drop probability between RED classes for any specific queue level, but I wouldn't know how to compute an overall relative drop balance.

BTW, I generally recommend against using (W)RED unless you're a QoS expert, especially in RED mode.  This particular QoS technology is, surprisingly, very difficult to obtain "ideal" results; and the Cisco implementation isn't, I believe, as "good" as some later RED variants.  (Also BTW, the history of RED's later variants, is interesting as the attempt to "fix" or "improve" RED's hoped for effectiveness.)

You might also find WRED helpful.  NB: I consider some of the information in the reference, inaccurate, but the WRED diagrams and explanations, I believe, better show what I tried to described, above.

[AlejandroJuarez03280]

Thank you for the information, now is more clear to understand the command output