QoS Question

[osvaldo Alcala Perez]

HI community, im having issues to understand why this QoS is behaving like this.

router is 4331/k9

my understanding is since the main Policy(parent) is CBWFQ it assigned up 50% and 40% to the class in the main. then the remaining BW is assigned to the class class-default based on this reamining bw, it is how the class is calculated? 

issues is: when the traffic CMV is running it always gets 50mbps no matter what.     all my email which is in COS 3 slows down really bad.. even when there is another 50Mbps available in the circuit...

how is BW really being calculated  for each parent class and class default ? 

maybe ???\

total bw available 100mbps(95Mbps due to shaping)

50mbps to the CMV traffic class

50Mbps remaining to the class class-default  

priority 30% of 50mbps = 15mbps 

Real remaining BW 35mbps?

20% to my COS3    30mbps * 20% = 7mbps available for the COS3 class?

interface GigabitEthernet0/0/0
description link to PE

bw 100Mbps
service-policy output SHAPE
output omitted

policy-map SHAPE
class HOPOFF
shape average percent 40
class CMV
shape average percent 50
class class-default
shape average percent 95
service-policy EGRESS_QUEUING

class COS1
priority percent 30
class COS2V
bandwidth remaining percent 25
fair-queue
police rate percent 25
conform-action set-dscp-transmit af41
exceed-action set-dscp-transmit af42
class COS2
bandwidth remaining percent 25
fair-queue
random-detect dscp-based
police rate percent 25
conform-action set-dscp-transmit af31
exceed-action set-dscp-transmit af32
class COS3
bandwidth remaining percent 25
fair-queue
random-detect dscp-based
police rate percent 25
conform-action set-dscp-transmit af21
exceed-action set-dscp-transmit af22

[paul driver]

Hello

looking at your qos policy’s  they don’t seem correct to me

you have shaping and policing applied to the same class maps and also enable wred for early random drops.

Your parent policy-map shape doesn’t seem to relate to any class-maps you’ve posted- i would have expected to see a single shape rate on this policy-map class-default relating to your Cir which would be

around  9216000

As for you child policy-map I would again expected to see all policing removed and just LLQ for the class COS1 and the other class-maps just having bandwidth remaining % with the addition of Fair queuing applied to the class class-default 

Apologies for the format of this post I am currently traveling and using a phone

[osvaldo Alcala Perez]

here are the class-map for the parent class

class-map match-any CMV
match access-group name CMV

class-map match-any HOPOFF
match dscp af13

[osvaldo Alcala Perez]

also the here is th child policy my bad . ;)

policy-map EGRESS_QUEUING
class NMC
bandwidth remaining percent 1
class COS1
priority percent 30
class COS2V
bandwidth remaining percent 25
fair-queue
police rate percent 25
conform-action set-dscp-transmit af41
exceed-action set-dscp-transmit af42
class COS2
bandwidth remaining percent 25
fair-queue
random-detect dscp-based
police rate percent 25
conform-action set-dscp-transmit af31
exceed-action set-dscp-transmit af32
class COS3
bandwidth remaining percent 25
fair-queue
random-detect dscp-based
police rate percent 25
conform-action set-dscp-transmit af21
exceed-action set-dscp-transmit af22

[paul driver]

Hello 

below is a example of what I have stated  previously however as I see Joseph has now joined this post I am sure you receive a better understanding from his reply’s 

Example:

Policy-map CHILD

class COS1
priority percent 30

class COS2V
bandwidth remaining percent 25

class COS2
bandwidth remaining percent 25

class COS3
bandwidth remaining percent 25

class class-default

bandwidth remaining percent 100

Fair queue

Policy-map PARENT

class class-default

shape-average 9216000

service-policy CHILD

[paul driver]

Policy-map CHILD

class COS1
priority percent 30

class COS2V
bandwidth remaining percent 25

class COS2
bandwidth remaining percent 25

class COS3
bandwidth remaining percent 25

class class-default

bandwidth remaining percent 100

Fair queue

Policy-map PARENT

class class-default

shape-average 9216000

service-policy CHILD

[Joseph W. Doherty]

Hmm, we could dissect your existing QoS commands but perhaps it would be better if you describe how you want your QoS policy to function (along with what the available bandwidth is).

[osvaldo Alcala Perez]

Joseph
thanks for reply. would you be kind and explain what is this doing. that is what im trying to accomplish with this post here...
may understanding is
main parent policy is CBWFQ and it is assigned with the 50% and 40% no matter what..with the shape average.
and the class class default do its BW assigments based on the remaining 10%?

again im trying to understand this...

[osvaldo Alcala Perez]

Available BW = 100mbps

[Joseph W. Doherty]

I'm not 100% sure. because, for starters, your interface policy, SHAPE, doesn't explicitly assign bandwidth to its three classes. I would assume each class gets a minimum guarantee of 1/3, but again, I unsure what Cisco will do (it's the kind of thing that might change between IOS versions).

Assuming SHAPE's classes (each) do have an implicit 1/3 minimum, then each SHAPE class has a percentage based shaper. Class HOPOFF will be shaped (limited) to 40% of the bandwidth, class CMV will be limited to 50% of the bandwidth and class-default will be limited to 95%. (It's unclear why the classes are being shaped unless there's some form of service agreement that "customers" are limited to their maximum bandwidth they might obtain. It's even more unclear why class-default is being shaped at 95%, as it's really so important to preclude using all the remaining [5%] of bandwidth?)

Continue, class-default also has an subordinate policy-map, EGRESS_QUEUING. This policy has multiple classes, which define how the parent policy's class-default queued traffic will be treated. 100% of remaining bandwidth hasn't been defined, so again, I'm unsure how what bandwidth ratio the implicit class-default will obtain. The other classes might actually have minimum bandwidth remaining percentages higher than what's configured. Some of the classes use both FQ and WRED. (Unless you're a real QoS expert, I suggest avoiding using WRED [especially when using FQ].)

Several classes have a policer which will "mark down" "overrate" traffic. This policy also has a LLQ defined, but since there are other parent policy classes, this LLQ packet's could be impacted by them. In theory, the other two classes are limited to using only up to 90% of the link capacity, but if these LLQ's packets are truly latency and/or jitter sensitive, the "norm" of not having LLQ "below" other parent polices should be followed.

"issues is: when the traffic CMV is running it always gets 50mbps no matter what."

That possibly (remember I mentioned SHAPE didn't define bandwidth allocations) shouldn't always be true. However, it may be true when there is CMV (at 50 Mbps) and only class-default (as [possibly - again undefined bandwidths] with just the two classes, each will get about 50%. Computing what a class like COS3 might obtain, depends on what else is happening. (Understanding, what class bandwidth statements actually "do" is set up dequeuing "weights".)