Reference Guide

All port queues are allocated with reserved buffers and when the reserved buffers are consumed, each queue starts using the
shared buffer from the default pool.
The reserved buffer per queue is 1664 bytes for the speed of 10G, 25G, 40G, 50G, and 100G. The default dynamic shared buffer
threshold value is 8.
Configure queue buffer settings
1. Create queuing type class-map to match the queue.
OS10(config)# class-map type queuing q1
OS10(config-cmap-queuing)# match queue 1
2. Create queuing type policy-map to define the actions for queues, like buffer configuration and thresholds.
OS10(config)# policy-map type queuing q-buffer
OS10(config-pmap-queuing)# class q1
OS10(config-pmap-c-que)# queue-limit queue-len 200 thresh-mode dynamic 5
Congestion avoidance
Congestion avoidance is a predictive method to anticipate and take necessary action to avoid congestion. The following are the
different mechanisms used to avoid congestion:
●
Tail drop—Packets are buffered at traffic queues. When the buffers are exhausted or reach the configured threshold,
excess packets are dropped. By default, OS10 uses tail drop for congestion avoidance.
● Random early detection (RED)—In tail drop mechanism, different flows are not considered in buffer utilization. When
multiple hosts start retransmission, tail drop causes TCP global re-synchronization. Instead of waiting for the queue to get
filled up completely, RED starts dropping excess packets with a certain drop-probability, when the average queue length
exceeds the configured minimum threshold. The early drop ensures that only some of TCP sources slow down, which avoids
global TCP re-synchronization.
● Weighted random early detection (WRED)—This mechanism allows different drop-probabilities and thresholds for each
color (red, yellow, green) of traffic. You can configure the drop characteristics for 3 different flows by assigning the
colors to the flow. You can assign colors to a particular flow or traffic by various methods, like ingress policing, qos input
policy-maps, and so on.
● Explicit congestion notification (ECN)—This is an extension of WRED. Instead of dropping the packets when the
average queue length crosses the minimum threshold values, ECN marks the Congestion Experienced (CE) bit of ECN field
in a packet as ECN-capable traffic (ECT).
Configure congestion avoidance
1. Configure WRED profile in the CONFIGURATION mode.
OS10(config)# wred wred_prof_1
2. Configure WRED threshold parameters for different colors in the WRED CONFIGURATION mode.
OS10(config-wred)# random-detect color yellow minimum-threshold 100 maximum-threshold
300 drop-probability 40
3. Configure the exponential weight value for the WRED profile in the WRED CONFIGURATION mode.
OS10(config-wred)# random-detect weight 4
4. Enable ECN.
OS10(config-wred)# random-detect ecn
5. Enable WRED/ECN on a queue.
OS10(config)# class-map type queuing c1
OS10(config-cmap-queuing)# match queue 2
OS10(config-cmap-queuing)# exit
OS10(config)# policy-map type queuing p1
OS10(config-pmap-queuing)# class c1
OS10(config-pmap-c-que)# random-detect wred_prof_1
Quality of service
643