14-7
Ethernet Card Software Feature and Configuration Guide, R7.2
January 2009
Chapter 14 Configuring Quality of Service
Scheduling
WDRR extends the quantum idea from the DRR to provide weighted throughput for each queue.
Different queues have different weights, and the quantum assigned to each queue in its round is
proportional to the relative weight of the queue among all t he queues serviced by that scheduler.
Weights are assigned to each queue as a result of the service provisioning process. When coupled with
policing and policy mapping provisioning, these weights and the WDRR scheduling process ensure that
QoS commitments are provided to each service flow.
Figure 14-5 illustrates the ML-Series card’s queuing and scheduling.
Figure 14-5 Queuing and Scheduling Model
The weighting structure allows traffic to be scheduled at 1/2048 of t he port rate. This equates to
approximately 488 kbps for traffic exiting a Gigabit Ethernet port, approximately 293 kbps for traffic
exiting an OC-12c port, and approximately 49 kbps for traffic exiting a FastEthernet port.
The unicast queues are created as the output service policy implementation on the egress ports. Each
unicast queue is assigned with a committed bandwidth and the weight of th e queue is determined by the
normalization of committed bandwidth of all defined unicast queues for that port. The traffic beyond the
committed bandwidth on any queue is treated by the scheduler according to the relative weight of the
queue.
The LLQ is created as the output service policy implementation on the egress ports. Each LLQ queue is
assigned with a committed bandwidth of 100 percent and is served with lower latency. To limit the
bandwidth usage by the LLQ, a strict policer needs to be implement ed on the ingress for the LLQ traffic
classes.
The DE allows some packets to be treated as committed and some as discard-eligible on the scheduler.
For Ethernet frames, the CoS (IEEE 802.1p) bits are used to identify committed and discard eligible
packets, where the RPR-CoS and the DE bits are used for Cisco pr oprietary RPR traffic. When
congestion occurs and a queue begins to fill, the DE packets hit a lower tail-drop threshold than the
committed packets. Committed packets are not dropped until the total committed load exceeds the
interface output. The tail-drop thresholds adjust dynamically in the card to maximize use of the shared
buffer pool while guaranteeing fairness under all conditions.
96502
Low Latency Queues
Weighted
Deficit
Round
Robin
Unicast Queues
Weighted
Deficit
Round
Robin
Weighted
Deficit
Round
Robin
Multi/Broadcast Queues
Weighted
Deficit
Round
Robin
Queues assigned by
"Priority" Command for
associated Classification
for Scheduling
Queues assigned by
"Bandwith" Command for
associated Classification
for Scheduling
Queues automatically
assigned