bandwidth percent and bandwidth remaining percent. WTD controls when queued packets are dropped, while queue buffers determine how much burst traffic can be held.If users report output drops, voice delay, or a QoS policy that appears ineffective, check traffic classification, queue scheduling weights, priority queues, WTD thresholds, queue buffers, and hardware policy installation separately.
What “Weighted Round Robin” Means on Catalyst 9200/9300
A switch may have several traffic classes competing for the same egress interface. When the interface becomes congested, packets are placed into queues and the scheduler determines which queue is served next.
A higher scheduling weight gives a queue more opportunities to use available bandwidth. This is the concept many users describe as weighted round robin.
On Catalyst 9200 and 9300 platforms, the practical configuration is generally based on bandwidth percent, bandwidth remaining percent, bandwidth remaining ratio, priority queues, queue limits, and queue buffer allocation.
The exact commands and supported parameters depend on the IOS XE release and platform. Check the applicable software documentation before deploying a QoS policy on a Cisco Catalyst 9200 switch.
Queue Scheduling, Priority Queues, and WTD
Queue scheduling, priority treatment, and Weighted Tail Drop solve different problems. Increasing a queue weight will not correct an incorrect DSCP-to-queue mapping, and changing a WTD threshold will not create additional physical bandwidth.
| Mechanism | Main Purpose | What It Does Not Solve |
|---|---|---|
| Scheduling weight | Shares bandwidth between queues | Incorrect traffic classification |
| Priority queue | Gives critical traffic preferential treatment | Unlimited congestion |
| WTD | Drops traffic after queue thresholds are exceeded | Lack of physical bandwidth |
| Queue buffer ratio | Allocates buffer space for bursts | Guarantees a transmission rate |
| DSCP/COS mapping | Places traffic into queues | Controls queue service order |
WTD uses traffic markings to apply different queue thresholds. When a queue reaches the threshold associated with a traffic class, new packets may be dropped even though the physical interface remains operational.
Cisco's Catalyst 9300 QoS Configuration Guide describes bandwidth allocation, queue scheduling, priority treatment, and WTD behavior for Catalyst QoS policies.
How Bandwidth Weights Affect Egress Queues
Bandwidth Percent
bandwidth percent assigns a minimum bandwidth percentage to a class. If the configured percentages do not consume the entire available allocation, the remaining bandwidth may be shared by other bandwidth queues.
This method is useful when the policy needs to reserve a predictable portion of an egress interface for a traffic class.
Bandwidth Remaining Percent
bandwidth remaining percent is useful when a policy also contains a priority queue. It assigns a relative share of remaining bandwidth to non-priority queues.
For example, video, business data, and bulk traffic could be assigned relative values of 40, 40, and 20. Under contention, the video and business-data queues receive a larger share of remaining bandwidth than the bulk queue.
These values are weights, not guaranteed throughput figures. A queue may use more bandwidth when other queues are idle, depending on the policy and platform behavior.
Queue Buffer Allocation
Queue buffers are different from bandwidth weights. A queue may have sufficient scheduling weight but still drop packets during a short burst if its available buffer is exhausted.
- Backup and storage replication traffic
- Video bursts
- High-speed uplinks feeding lower-speed ports
- Multiple access ports converging on one egress interface
A Practical QoS Policy Example
The following example illustrates how traffic classes and remaining-bandwidth weights can be organized. It is not a universal production template. Confirm the supported syntax for the installed IOS XE release before applying it.
Before applying a similar policy, confirm that DSCP markings are trustworthy, the class-maps match the intended traffic, the interface supports the selected QoS features, and the policy does not mix incompatible bandwidth types.
A high-density Cisco Catalyst 9300L-48P-4X-E may aggregate more traffic than an access-layer Catalyst 9200. The congestion point may therefore occur on a 10G uplink, a 1G endpoint port, or a downstream port depending on the topology.
How to Verify Queue Behavior
Confirm the Policy Is Attached
Start with the interface and policy configuration. Confirm that the policy is attached in the correct direction and that the expected class counters are increasing.
Check Interface-Level Drops
High output drops usually point to egress congestion, but interface counters alone may not identify the affected queue or traffic class.
Check Queue Statistics
Look for queue-specific drops, queue occupancy, threshold counters, and differences between ingress and egress behavior. The exact command syntax can vary by IOS XE release.
Verify Policy Installation in Hardware
A policy can be accepted by the configuration parser but still fail to install correctly in hardware. For a stack interface, select the correct switch member.
A status such as VALID,SET_INHW indicates that the policy is understood by software and installed in hardware. An invalid state should be investigated before treating the policy as active.
Check Hardware Resources
For a stack, replace active with the appropriate switch number when necessary. A policy that fails to install may be affected by unsupported syntax, hardware resource limits, or IOS XE release restrictions.
Why QoS Queues Drop Packets
Voice or Video Traffic Is Delayed
- DSCP values are not preserved.
- Traffic is mapped to the default queue.
- The intended class-map does not match.
- The priority queue is not configured.
- The egress link is oversubscribed.
Start with the policy counters and class-map configuration before changing queue weights.
Bulk Traffic Consumes Most of the Link
Possible causes include bulk traffic entering a high-priority class, overly generous bandwidth weights, incorrect business-traffic classification, or a service policy applied in the wrong direction.
Output Drops Increase Despite Low Average Utilization
Average utilization can hide microbursts. Short bursts may fill a queue before interface-level average rates appear high.
- Review queue statistics.
- Check buffer allocation.
- Look for speed mismatches.
- Check storage, backup, and video traffic patterns.
- Identify multiple high-speed ports converging on one lower-speed egress port.
The QoS Policy Is Configured but Has No Effect
Check whether the service policy is attached, the interface direction is correct, the class-map is matching packets, the policy is installed in hardware, and the selected commands are supported by the IOS XE release.
WTD Causes Unexpected Packet Loss
Review which QoS label is assigned to the traffic, which threshold that label uses, whether the queue is receiving a sudden burst, and whether the queue buffer allocation is sufficient.
Raising a WTD threshold may reduce drops, but it can also allow one traffic class to consume more shared buffer space. Treat it as a design change rather than a universal fix.
Send the switch model, IOS XE release, interface role, policy-map, and queue-drop output.
Frequently Asked Questions
show policy-map interface and the FED policy target status.Final Takeaway
For Catalyst 9200 and 9300 switches, “weighted round robin” is best analyzed as a combination of queue scheduling, bandwidth weights, priority handling, WTD thresholds, and buffer allocation.
When troubleshooting congestion, follow this sequence:
- Confirm where traffic is classified.
- Verify that the policy is attached.
- Check queue and interface counters.
- Review scheduling weights and priority settings.
- Check WTD and buffer behavior.
- Confirm that the policy is installed in hardware.
- Check hardware resources and IOS XE compatibility.
A change to one queue weight will not solve a problem caused by incorrect classification, insufficient buffering, or a policy that never reached the ASIC.