Ipv6 QoS

If you already grasp QoS concepts for IPv4, IPv6 QoS is a piece of cake!

As with IPv4, IPv6 uses MQC (Modular QoS CLI)to configure Diffserv (Differentiated services) QoS.

IPv6 QoS is very similar to IPv4 QoS except for some details:

  • NBAR, the first version, doesn’t support IPv6.
  • cRTP (Compressed-RTP).
  • No way to match directly RTP.
  • CAR (Committed Access Rate) replaced by CB- Policing already in IPv4 and no need to keep supporting it in IPv6.
  • PQ/CQ replaced by MQC (Modular QoS CLI).
  • IPv6 supports only named ACL.
  • Layer2 (802.1q) commands works only with CEF- Switched ports not with process- switched nor router originated traffic.

The following is the topology used to deploy IPv6 QoS, no IPv4 addressing scheme. The serial link between the two routers is the bottleneck of the network where QoS is needed.

Figure1: Topology

Classification & Marking

The first and the most crucial step in deploying QoS is classification of traffic.

In this step you need to:

  • identify various applications and protocols running on your network.
  • understand the application behavior with respect to the available network resources.
  • identify the mission critical and non-critical application.
  • Categorize the applications and protocols in different classes of service accordingly.

The classification is based on packet native classifiers like:

  • source/destination IPv6 addresses, IP protocol and source/destination ports.
  • precedence and dscp.
  • source/destination MAC.
  • TCP/IP header parameters (packet length…).
  • IPv6 specific classifiers (not currently used).
  • IPv6 (traffic class) used in the same way as IPv4 (ToS).

IPv4 can take advantage of NBAR, very useful to automatically recognize applications and provides statistics about bandwidth utilization. Without NBAR, you will need to manually determine which classifiers define the application you want QoS to handle. Unfortunately NBAR doesn’t support IPv6, NBAR2 does.

For IPv6 traffic, you can use other tools such “Netflow” or any traffic analyzer software for more granular inspection, then build IPv6 ACLs matching the relevant classifiers with the relevant values.

table1: Application classification and marking

ACL name








Src port



Dst port





ftp (21)





ftp-data (20)








Table1 summarizes the applications used in the lab for demonstration purpose.

Table2: Application classification and marking


Bandwidth allocated

Flow direction

traffic classifiers



unicast streaming

700 kbps

From HostB to HostA

dest IPv6=2001:a:a:a::a



protocol I =Pv6

dest port 1234

FTP download

30 kbps

From HostB to HostA

src port 21 (control)



protocol I =Pv6

src port 20 (data)

scavenger appli

video streaming

150 kbps

From HostB to Host A

src port …

Generally dscp “ef” is reserved for VoIP which requires the most stringent QoS, in this lab we use the dscp marking  just to check at the destination host (hostA) whether the classification works.

The end-to-end model used to test IPv6 QoS is depicted in Figure2.

Figure2: End-to-End QoS model

Congestion Management & avoidance

For the purpose of the lab, the unicast streaming application is given the highest priority and it is supposed to have stringent bandwidth, latency, delay and jitter requirements, LLQ (Low Latency Queuing) is the most appropriate queuing mechanism for such applications.

The FTP traffic is considered critical with a minimum of  30kbps of bandwidth guaranteed .

Any other traffic, default-class is considered “scavenger” and will have no privilege during congestion.

Each application is being allocated the needed bandwidth to perform correctly.

Table3: Classes and bandwidth allocation

Class Bandwidth reserved Queue DSCP Priority
MatchUStream 700 kbps LLQ af41 High
MatchFTP 30 kbps CBWFQ af21 Medium
class-default no guarantee WFQ 0 Low
policy-map QoS_Policy
  class MatchUStream
   set dscp ef

   priority 700

  class MatchFTP

   set dscp af41

   bandwidth 30

  class class-default


   set dscp default

Figure3 and 4 show a summary of general QoS mechanisms and queuing system types.

Figure3: Software and Hardware queuing systems

Figure4: QoS mechanisms


ipv6 access-list FTP
permit tcp host 2001:B:B:B::B eq ftp any
permit tcp host 2001:B:B:B::B eq ftp-data any


ipv6 access-list UStream

sequence 20 permit udp any any eq 1234


class-map match-all MatchFTP

  match protocol ipv6

  match access-group name FTP

class-map match-all MatchUStream

  match protocol ipv6

  match access-group name UStream


RouterB(config-pmap-c)#do show policy-map int s1/0
  Service-policy output: QoS_Policy

    Class-map: MatchUStream (match-all)

      23625 packets, 32602500 bytes

      30 second offered rate 538000 bps, drop rate 0 bps

      Match: protocol ipv6

      Match: access-group name UStream

      QoS Set

        dscp ef

          Packets marked 23624


        Strict Priority

        Output Queue: Conversation 264

        Bandwidth 700 (kbps) Burst 17500 (Bytes)

        (pkts matched/bytes matched) 1455/2007900

        (total drops/bytes drops) 1/1380

    Class-map: MatchFTP (match-all)

      5886 packets, 8192512 bytes

      30 second offered rate 135000 bps, drop rate 0 bps

      Match: protocol ipv6

      Match: access-group name FTP

      QoS Set

        dscp af41

          Packets marked 5929


        Output Queue: Conversation 265

        Bandwidth 30 (kbps) Max Threshold 64 (packets)

        (pkts matched/bytes matched) 3486/4784640

        (depth/total drops/no-buffer drops) 0/0/0

    Class-map: class-default (match-any)

      105 packets, 8292 bytes

      30 second offered rate 0 bps, drop rate 0 bps

      Match: any


        Flow Based Fair Queueing

        Maximum Number of Hashed Queues 256

        (total queued/total drops/no-buffer drops) 0/0/0

      QoS Set

        dscp default

          Packets marked 50


RouterB(config-pmap-c)#do sh int s1/0
Serial1/0 is up, line protocol is up
  Hardware is M4T

  MTU 1500 bytes, BW 1024 Kbit, DLY 20000 usec,

     reliability 255/255, txload 165/255, rxload 1/255

  Encapsulation HDLC, crc 16, loopback not set

  Keepalive set (10 sec)

  Restart-Delay is 0 secs

  Last input 00:00:07, output 00:00:00, output hang never

  Last clearing of “show interface” counters 00:08:49

  Input queue: 0/75/0/0 (size/max/drops/flushes); Total output drops: 21

  Queueing strategy: weighted fair

  Output queue: 0/1000/64/1 (size/max total/threshold/drops)

     Conversations  0/3/256 (active/max active/max total)

     Reserved Conversations 1/1 (allocated/max allocated)

     Available Bandwidth 38 kilobits/sec

  30 second input rate 4000 bits/sec, 8 packets/sec

  30 second output rate 666000 bits/sec, 59 packets/sec

     4066 packets input, 261676 bytes, 0 no buffer

     Received 61 broadcasts, 0 runts, 0 giants, 0 throttles

     0 input errors, 0 CRC, 0 frame, 0 overrun, 0 ignored, 0 abort

     32082 packets output, 44217146 bytes, 0 underruns

     0 output errors, 0 collisions, 0 interface resets

     0 output buffer failures, 0 output buffers swapped out

     0 carrier transitions     DCD=up  DSR=up  DTR=up  RTS=up  CTS=up


Shaping & Policing

Shaping and policing work exactly as in IPv4.


  • Applied inbound and outbound.
  • Drops non conforming traffic.
  • More efficient in term of memory utilization.
  • Drops packets more often, therefore more TCP retransmission.


  • Applied only outbound.
  • Queues excess traffic.
  • Less efficient because of additional queuing, but less dropping  (only when congestion occurs).
  • Causes variable delay (jitter) and increases buffer utilization, therefore more delays.

Figure5 shows the mechanism of token bucket used in shaping and policing.

Figure5: shaping and policing

Here is how the FTP traffic diagram looks like before any shaping or policing:

Figure6: FTP before shaping and policing

The following figurex shows different behaviors based on three configurations of shaping and traffic:

Figure7: FTP with shaping and traffic

The first part of the graph corresponds to FTP traffic with just a configured guaranteed bandwidth of 30kbps.

policy-map QoS_Policy
  class MatchFTP
   bandwidth 30

In the second part of the graph, a high limit is set for FTP class using policing at 100kbps, you can note that this results in a frequent TCP global synchronization, a TCP protocol behavior when congestion occurs somewhere in the path to the destination, as long as the congestion exists the source continues to receive requests to decrease the sending rate back from zero and so on, hence the form of the graph (repeated short bursts from the bottom to the maximum).

policy-map QoS_Policy
  class MatchFTP
   bandwidth 30

   police 100000

The third part of the graph represents the result of using shaping instead of policing, more optimal use of the bandwidth. Instead of dropping the TCP traffic and causing global synchronization, the exceed packets are queued for a certain amount of time and then sent, hence the higher used average bandwidth.

policy-map QoS_Policy
  class MatchFTP
   bandwidth 30

   shape average 100000

RouterB#sh policy-map int s1/0

    Class-map: MatchFTP (match-all)

      32045 packets, 47038153 bytes

      30 second offered rate 99000 bps, drop rate 0 bps

      Match: protocol ipv6

      Match: access-group name FTP

      QoS Set

        dscp af41

          Packets marked 32074


        Output Queue: Conversation 265

        Bandwidth 30 (kbps) Max Threshold 64 (packets)

        (pkts matched/bytes matched) 17595/26364666

        (depth/total drops/no-buffer drops) 0/0/0

      Traffic Shaping

           Target/Average   Byte   Sustain   Excess    Interval  Increment

             Rate           Limit  bits/int  bits/int  (ms)      (bytes)

           100000/100000    2000   8000      8000      80        1000

        Adapt  Queue     Packets   Bytes     Packets   Bytes     Shaping

        Active Depth                         Delayed   Delayed   Active

        –      11        17134     25746208  17096     25689056  yes


Make sure you understand first IPv4 QoS, especially the difference between shaping and policing and their impact on your own applications.


%d bloggers like this: