Switch QoS

4500 Platform

In this blog, we will take a look at the QoS capabilities of the 4500 series switch. Quality of Service (QoS) processing on the 4500 series switch is done at the supervisor module. QoS support requires the Supervisor 2+ or later supervisor module with each supervisor supporting different capabilities.

QoS processing is not turned on by default. QoS is turned on by using the following global IOS command:

Cat4500(config)#qos

There are many configuration commands on the 4500 that are similar to commands used on other native IOS switches, but command syntax never includes the mls prefix (multi-layer switching). The 3550/3560/2960/2970/3750 switches require the mls qos command to globally enable qos processing.

The Catalyst 4500 supports a unique congestion avoidance algorithm called DBL (dynamic buffer limiting). The 2950/3550/6500 switches support the WRED congestion avoidance algorithm that works on a per packet basis, while DBL works at the flow level. A flow is identified by the VLAN combined with the source and destination layer 3 (IP Address) and layer 4 (TCP/UDP port) information. DBL allows behaving flows to be preserved while misbehaving aggressive flows are punished as a queue exceeds a configurable threshold. DBL is effective with non-adaptive flows (NAF), while WRED is not. Non-adaptive flows do not respond to packet loss by lowering their window sizes like TCP based traffic. Most UDP based applications (voice and/or video) are like packet generators because they do not throttle back their rates in any way when packets are lost. DBL must be configured globally with the following command syntax:

Cat4500(config)#qos dbl

The DBL capabilities of the switch also have the ability to mark the explicit congestion notification (ECN) bits of the type of service (ToS) byte in the IP header. RFC3168 introduced ECN capabilities into the type of service byte so routers and switches could mark packets as congestion experienced when queues exceed a threshold. The source or destination TCP/IP stack will then lower their window size to react to the congestion on the network. Congestion avoidance algorithm are utilized during periods of congestion, which begs the question: Why are the mechanisms called congestion avoidance if they kick in during periods of congestion? The algorithms avoid full queue congestion which would result in tail drop where all traffic is dropped regardless of priority. The ECN capabilities are enabled with the following command:

Cat4500(config)#qos dbl exceed-action ecn

The global DBL commands shown above enable the use of the dbl command in individual classes at the policymap-class configuration level.

The 4500 switch supports CoS and DSCP trust boundary configurations, but all QoS internal processing and transmit queue mapping is done based on the DSCP. Classification and marking is supported with the use of a service-policy, but only certain line card modules can have a service-policy directly attached to a port. Classification and marking policies are applied to vlans on the 4500 by default, but newer versions of IOS on the 4500 allow per vlan, per port QoS. An example of per port, per vlan QoS is shown below:

Cat4500(config)#interface gigabitethernet 5/1
Cat4500(config-if)#vlan-range 5-10
Cat4500(config-if-vlan-range)#service-policy input classify-mark

In the above example, the policy-map named classify-mark is applied to input traffic on vlan 5 through 10 on gigabitethernet 5/1 which is an 802.1q trunk to another switch. Other VLANs on the interface do not have this policy-map attached to their traffic.

The 4500 queue architecture includes 4 queues in which DSCP values are mapped to the individual queues. Newer line card modules like the Supervisor Engine 6-E have eight transmit queues in which traffic can be mapped. We will take a look at the DSCP mapping used in the four queue architectures. The queue mapping is as follows:

DSCP Range Queue number
0-15 1
16-31 2
32-48 3
49-63 4

Queue 3 can be configured as a priority queue and shaped to 1Mbps with the following command syntax:

Cat4500(config)#interface gigabitethernet 5/1
Cat4500(config-if)#tx-queue 3
Cat4500(config-if-tx-queue)#priority high
Cat4500(config-if-tx-queue)#shape percent 33
Cat4500(config-if-tx-queue)#bandwidth percent 33

The priority queue shaping option is similar to shaped round robin (SRR) on the 2960/2970/3560/3750 switches. Each queue can be assigned a bandwidth parameter to control the weighting of each queue’s processing during periods of congestion. Only the bandwidth percentage of queue 3 is shown in the example above.

Ingress and egress policers are supported on the 4500 switch. The police command on the 4500 switch supports speed increment keywords (kbps and mbps) to indicate the rates at which traffic should be policed. Other switches require rates to be configured in bits per second (bps) which results in long, cumbersome numbers.

The 4500 QoS commands and capabilities differ by supervisor module. For more information on the 4500 QoS capabilities, check out the reference links below.

REFERENCES
Quality of Service on Cisco Catalyst 4500 Series
http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps4324/prod_w...

Quality of Service on the Cisco Catalyst 4500 Supervisor Engine 6-E
http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps4324/white_...

Configuring Quality of Service – 12.2(52)SG
http://www.cisco.com/en/US/docs/switches/lan/catalyst4500/12.2/52sg/conf...

QoS SRND - Catalyst 4500 Supervisor II+/III/IV/V—QoS Considerations and Design
http://www.cisco.com/en/US/docs/solutions/Enterprise/WAN_and_MAN/QoS_SRN...

• Careers
• Cisco
• DBL UBRL 4507-R 4507 ECN NAF WRED