Specifications

White Paper

4. The DSCP can be set for the frame using a DSCP default value typically assigned though an Access Control

List (ACL) entry.

After a DSCP value is assigned to the frame, policing (rate limiting) is applied should a policing configuration exist.

Policing will limit the flow of data through the PFC by dropping or marking down traffic that is out of profile. Out of

Profile is a term used to indicate that traffic has exceeded a limit defined by the administrator as the amount of bits

per second the PFC will send. Out of profile traffic can be dropped, or, the data can still be sent but CoS value is

marked down. The PFC1 and PFC2 currently only support input policing (rate limiting). The PFC3 supports input and

output policing. The output-policing feature of the PFC3 applies to routed (layer 3) ports or VLAN interfaces (switch

SVI interface—this is discussed in more detail later in the paper).

The PFC will then pass the frame to the egress port for processing. At this point, a rewrite process is invoked to

modify the CoS values in the frame and the ToS value in the IPv4 header. Prior to passing the frame to the port

ASIC, the PFC will derive the CoS based on internal DSCP. The port ASIC will then use the CoS value passed to it

to place the frame into the appropriate queue. While the frame is in the queue, the port ASIC will monitor the buffers

and implement WRED to avoid the buffers from overflowing. A Round Robin scheduling algorithm is then used to

schedule and transmit frames from the egress port

Each of the sections below will explore this flow in more detail giving configuration examples for each of the steps

described above.

6. Queues, Buffers, Thresholds and Mappings

Before QoS configuration is described in detail, certain terms must be explained further to ensure the reader fully

understands the QoS configuration capabilities of the switch.

6.1 Queues

Each port in the switch has a series of input and output queues that are used as temporary storage areas for data.

Catalyst 6500 line cards implement different number of queues for each port. The queues are usually implemented

in hardware ASIC’s for each port. This is different to routers where the queues are virtualised by the software. On the

first generation Catalyst 6500 line cards, the typical queue configuration included one input queue and two output

queues. Later line cards use enhanced ASIC’s which incorporated additional queues. One innovation included

support for a special strict priority (SP) queue, which is ideal for storing latency sensitive traffic like Voice over IP

(VoIP). Data in this queue is serviced in a strict priority fashion. That is, if a frame arrives in the SP queue,

scheduling and transmission of frames from the lower queues is ceased in order to process the frame in the strict

priority queue. Only when the SP queue is empty will scheduling of packets from the lower queue(s) recommence.

When a frame arrives at an ingress port and congestion is present, it will be placed into a queue. The decision

behind which queue the frame is placed in is determined by the CoS value in the Ethernet header of the incoming

frame.

On egress, a scheduling algorithm will be employed to empty the transmit (output) queue. There are a number of

Round Robin techniques available (dependent on the hardware and software in use) to perform this function. These

Round Robin techniques include Weighted Round Robin (WRR), Deficit Weighted Round Robin (DWRR) and

Shaped Round Robin (SRR). While each of these are explored later, in the case of WRR and DWRR, each queue

uses a weighting to dictate how much data will be emptied from the queue before moving onto the next queue. For

SRR, a rate (or limit) is applied on a per queue basis and this dictates the upper limit of bandwidth that can be used

by this queue.