QOS bandwidth allocation

[mustansirt]

If I have the below policing configured on the input of the interface of the ISP port 

class realtime_dscp

  police cir 1952000 bc 242000 be 486000 conform-action set-mpls-exp-imposition-transmit 5 exceed-action drop

class time_critical_dscp

  police cir 976000 bc 120000 be 243000 conform-action set-mpls-exp-imposition-transmit 4 exceed-action set-mpls-exp-imposition-transmit 0

class besteffort_dscp

  police cir 936000 bc 116000 be 233000 conform-action transmit  exceed-action transmit

so if there is no real time traffic and only best effort traffic then the best effort traffic can use the available bandwidth of  real-time.

2nd question 

if the above is true, then as the best effort traffic is already using the available bandwidth then if real time traffic starts to come in then how will it be treated. Will it have to fight for the bandwidth with other class or real time will get the bandwidth which is specified in the police cir i.e. 1952000.

[Joseph W. Doherty]

Just wanted to know Like Cisco has an ideal QOS design for enterprise.

Laugh, I would debate what Cisco suggest is an ideal QoS design.  BTW, that's not knocking what Cisco suggests, it's that Qos design if often influenced by the service needs of the traffic.

do they suggest one for Service provider i.e how much should be reserved for real-time and for data and best effort.

I don't recall seeing any Cisco QoS suggestions for SPs.  In my experience, SPs generally just do BE for all traffic.  If a customer has bandwidth congestion issues, SPs will often suggest obtaining more bandwidth or an additional circuit rather than managing bandwidth via QoS.  However, they do sometimes offer some QoS support, also sometimes at additional cost.

[mustansirt]

Thanks joseph for the reply. 

As ios XE has below command for giving priority bandwidth 

policy XYZ

class voice

priority 256

how can i replicate this in IOS XR.

[Joseph W. Doherty]

Sorry, I haven't worked with XR.

[mustansirt]

Dear Joseph,

I currently have below policy-map configured at egress of the wan

policy-map p_egress_p

class real_time

  police rate percent 25

   exceed-action drop

!

class business

  bandwidth percent 20

  random-detect 232 packets 774 packets

!

class mgmt

  bandwidth percent 2

!

class routing

  bandwidth percent 2

!

class time_critical

  bandwidth percent 15

  random-detect 99 packets 332 packets

!

class besteffort

  bandwidth percent 35

!

class scavenger

  bandwidth percent 1

  random-detect 7 packets 22 packets

!

class class-default

!

end-policy-map

1) does the high end ASR have maximum reserved bandwidth limitation of 75 % configured by default on their interface.

2) question on each of above class.

a)

class real_time :- It's only policing and not prioritizing the real-time traffic.(right)

b)the classes time-critical , business and scavenger have random detect configured on each of their classes.so each of these classes will have WRED enabled under their classes to drop packtes before congestion,

can you explain what exactly the random detect command above is doing, because it's configured based on packets.

c) if we have priority under the real-time class then if there are packets coming for each of the above class,  the real-time packet will be send first and all the packets of other remaining classes will compete with each other .am i right or there will be a sequence in which they will be treated in the outbound queue based on their DSCP,precedence or COS value.

[Joseph W. Doherty]

#1 Not as far as I know.  I believe all CBWFQ, since the introduction of HQF, removed the implicit max 75% reservation.

#2 a: I'm not sure what it will do any you didn't provide it a priority or bandwidth statement.

#2 b: I'm not familiar with that variant of syntax - this on an ASR?

Suspect it's setting the min and max thresholds.  If so, the chance of dropping newly added packets to the queue begins at min and hits a max defined percentage at max (it's also 100% at max plus one).    What the drop possibility percentage is between min and max can be computed by for each packet depth value.  (Basically, you draw a line starting with x=min and y=0 at min threshold, to x=max and y=max drop percentage at max threshold.)

WRED's queue depth is a moving average, not the actual current queue depth.

Personally, I recommend against using WRED unless you're a real QoS wizard.

#2 c: Priority packets are queued to a LLQ, where they are dequeued before any other non-LLQ packets.  To keep them from taking all bandwidth from non-LLQ, the priority command also sets an implicit policer.

Non-LLQ packets are dequeue relative to each other based on their bandwidth rations.  For example, if class A is 50% and class B 25% or class A is 6% and class B is 3%, class A traffic should be able to obtain twice as much bandwidth as class B.

With most CBWFQ, DSCP or CoS values only usually help direct traffic to specific classes.  Once it's in a class, the class configuration determines classes egress bandwidth.

For example I might have a class map that matches DSCP CS1 or FTP which directs that traffic to a low priority class.  The class dequeuing doesn't care how the traffic got to the class.

[mustansirt]

Dear joseph 

what i wanted to ask is when packets are queued at the egress queue are they queued in the order of their marking i.e Ef , AF41 ,AF31....0. or 1st will be the priority packets and then rest of the packets irrespective of their marking.

[Joseph W. Doherty]

It's up to your policy, i.e. it could use ToS markings or not (or ToS marking and other packet attributes) to decide what queue packets are queued to and in what sequence packets are dequeued.

[mustansirt]

In case of CBWFQ  the priority packet is sent first , then the rest of the packet which will be sent after priority packet i.e. the next packet in sequence will be according to the percentage assigned to each class like class with more percentage bandwidth will be sent more often than the class with less percentage

or depending upon the cos marking of the packet , will the next packet after priority packet will be sent like COS 5(EF)than COS4(AF4x)and so on and last best effort.

[Joseph W. Doherty]

Any/all LLQ packets will be dequeued before any other non-LLQ packets.  (NB: LLQ packets also might be dropped by implicit class policer too.)

Other classes are dequeued based on their allocated share of bandwidth.

Again, ToS may be part of the criteria of which determines what class a packet is directed to, but with post-HQF CBWFQ, once a packet gets to an egress queue, ToS no longer matters.  (Original WFQ did use ToS for weighting its queues.  Pre-HQF CBWFQ, class-default fair-queue also used WFQ.)

PS:

BTW, this is now the 3rd time I've tried to answer your question.  Once again, ToS may matter determining what class a packet is directed to, but once packet gets to class, ToS usually doesn't matter for dequeuing prioritization.

[mustansirt]

Thanks Joseph for all your replies, just trying to get things cleared.

[Joseph W. Doherty]

Understood - and if still unclear - please let me know.