Specifications

White Paper

1. What is Layer 2 and Layer 3 QoS

While some may think of QoS in a switch is only about prioritising Ethernet frames, it is in fact much more. Layer 2

and Layer 3 QoS in the Catalyst 6500 entails the following:

1. Input Trust and Queue Scheduling: When the frame enters the switch port, it can be assigned to an ingress

port-based queue prior to being switched to an egress port. Different queues can be used for different traffic that

require different service levels, or where switch latency must be kept to a minimum. For instance, IP based

video and voice data requires low latency, so there may be a need to switch this data prior to switching other

data like FTP, web, email, telnet, etc. The Trust element of QoS determines whether to honor any existing

priority settings incorporated in the incoming frame.

2. Classification: The process of classification is one of inspecting different fields in the Ethernet Layer 2 header,

along with fields in the IP header (Layer 3) and the TCP/UDP header (Layer 4) to determine the level of service

that should be applied to the frame as it transits the switch.

3. Policing: Policing is the process of inspecting whether traffic on a given port or within a VLAN has exceeded a

pre defined rate. Should that traffic be out of profile (i.e. the traffic stream is in excess of the pre defined rate

limit), excess data can be either dropped or its priority value can be marked down.

4. Marking: The process of rewriting is the ability of the switch to modify the priority of the packet via the CoS in

the Ethernet header or the Type of Service bits in the IPv4 header

5. Output Queue Congestion Management and Scheduling: When the packet is ready to be switched to its next

hop destination, the switch will place the Ethernet frame into an appropriate outbound (egress) queue for

switching. The switch will perform buffer (congestion) management on this queue by monitoring the utilization.

Congestion management utilizes 4 different algorithm:

a. Random Early Discard (RED) algorithm whereby random frames are refused admission to the queue once a

threshold has been exceeded.

b. Weighted RED (WRED) is a derivative of RED whereby the frames priority values are inspected to determine

which frames will be dropped. When the buffers reach set thresholds, then (typically) lower priority frames are

dropped allowing the higher priority frames to enter the queue. Scheduling of the packet is performed by a round

robin (and possibly strict priority) scheduling technique that moves between the queues to forward the packets.

c. Deficit Weighted Round Robin (DWRR), this algorithm is similar to WRED but monitor utilization in the

transmit queues. If a queue uses excess bandwidth, the DWRR algorithm keeps track of this excess usage and

subtracts it from the queue’s next transmission bandwidth allocation. This results in a fairer use of bandwidth

over time.

d. Shaped Round Robin (SRR), with SRR, each queue gets a “SHARE”, which translates into a fraction of the

output time this queue can send data for.

This paper will explain in more detail each of the mechanisms above and how they relate to the Catalyst 6500.

2. Why The Need for QoS in a Switch

References to large capacity switching backplanes, millions of switched packets per second and non-blocking

switches are synonymous terms used with many switches today. Why then do we need QoS? The answer is simply

because of congestion.