The Common Misconception
“We don’t need QoS — we’ll just upgrad our links.”
That’s a common but dangerous mindset. QoS doesn’t exist to fix slow links — it exists to manage congestion intelligently when multiple high-speed flows compete for the same egress queue.
The Reality of Link Ratios
Upgrading bandwidth helps, but ratios matter more.
If your servers go from 1 Gbps → 10 Gbps and your uplinks go from 10 Gbps → 100 Gbps, the oversubscription ratio (1:10) hasn’t changed. Packets can still build up and drop when multiple sources transmit simultaneously.
The real challenge isn’t speed — it’s flow balance, buffering, and scheduling.
Understanding Cisco Platform Differences
Platform | ASIC | Buffering | QoS Type | Use Case |
---|---|---|---|---|
Catalyst 9000 | UADP | Shared or per-port (varies by model; some use VoQ) | MQC (Modular QoS CLI) | Campus & Access |
Nexus 9000 / ACI | Broadcom Trident / Jericho | Deep buffers with VoQ (depth varies significantly by model) | DCB, PFC, ECN | Data Center |
Cisco 8000 Series (Silicon One) | Q200 / Q100 / P100 | On-chip deep buffers + HBM | Hierarchical QoS (HQoS, depth varies by model) | WAN / Core / Edge |
Important Note: Buffer architectures and QoS capabilities vary significantly within each platform family. For example, a Catalyst 9300 differs from a 9500, and a Nexus 93180YC-FX has vastly different buffering than a 9736C-FX. Always consult platform-specific datasheets and the Cisco QoS design guide for your exact model and software version.
The Root of the Problem
Traffic bursts exceed buffer capacity → queues fill → packets drop → retransmissions begin → jitter and delay increase.
This happens even at 400 Gbps if buffers or QoS policies aren’t tuned. Different Cisco platforms handle this differently — so your QoS design must align with hardware architecture.
Practical Cisco Remediation Examples
A. Monitor Queues & Drops
Catalyst 9000
show interfaces counters errors show interfaces queue | include drops show platform hardware qos queue stats interface TenGigabitEthernet1/0/1
Nexus 9000
show interface Ethernet1/1 counters errors show queuing interface Ethernet1/1 show hardware qos queue stats interface Ethernet1/1
Cisco 8000 Series (IOS XR)
show qos interface HundredGigE0/0/0/0 show platform hardware qfp active statistics drop show platform hardware qfp active qos queue output
B. Enable QoS Globally (Where Required)
Nexus 9000
configure terminal feature qos feature queuing
Cisco 8000 (IOS XR)
configure qos enable
C. Configure QoS Classes & Policies
Catalyst 9000 (Access Layer)
class-map match-any VOICE match dscp ef class-map match-any VIDEO match dscp af41 ! policy-map CAMPUS-QOS class VOICE priority percent 20 class VIDEO bandwidth percent 30 class class-default fair-queue ! interface TenGigabitEthernet1/0/1 service-policy output CAMPUS-QOS
Nexus 9000 (Data Center)
policy-map type qos DC-QOS class type qos AI-TRAFFIC set qos-group 5 class type qos class-default set qos-group 0 ! system qos service-policy type qos input DC-QOS
Cisco 8000 (Core WAN)
policy-map CORE-QOS class VOICE priority level 1 class VIDEO bandwidth percent 30 class BULK bandwidth percent 10 class class-default fair-queue ! interface HundredGigE0/0/0/0 service-policy output CORE-QOS
D. Tune Buffers & Queues
Catalyst 9000
interface GigabitEthernet1/0/1 hold-queue 4096 out
Nexus 9000
hardware qos queue-policy QPOLICY-10G queue 1 bandwidth percent 20 queue 2 bandwidth percent 30 queue 3 bandwidth percent 50 ! interface Ethernet1/1 service-policy type queuing output QPOLICY-10G
Cisco 8000 (IOS XR)
policy-map QUEUE-TUNING class VOICE queue-limit 300 ms class VIDEO queue-limit 500 ms class BULK queue-limit 800 ms ! interface Bundle-Ether10 service-policy output QUEUE-TUNING
E. Implement PFC & ECN (For Lossless AI / HPC Traffic)
Nexus 9000
priority-flow-control mode on priority-flow-control priority 3 enable
Cisco 8000 (IOS XR)
policy-map LOSSLESS class HPC pause no-drop ecn ! system qos service-policy type network-qos LOSSLESS
F. Validate QoS Behavior
show policy-map interface show queuing interface show hardware qos queue stats all show qos interface statistics
Cisco Platform Design Summary
Platform | Design Focus | Key QoS Features | Typical Use |
---|---|---|---|
Catalyst 9000 | Campus / Access | MQC QoS, per-port queues, fair-queue | Edge/LAN |
Nexus 9000 / ACI | Data Center Fabric | VoQ, PFC, ECN, deep buffers | East–West traffic, AI fabrics |
Cisco 8000 Series | WAN / Core / Cloud Edge | Hierarchical QoS, Silicon One ASIC, adaptive congestion visibility | High-throughput backbones, WAN aggregation |
Hierarchical QoS (HQoS) on Cisco 8000 Series
Hierarchical QoS allows multiple levels of control:
- Level 1 (Parent Policy): Shape the aggregate bandwidth at the interface level.
- Level 2 (Child Policy): Assign bandwidth and priority to individual traffic classes.
- Level 3 (Grandchild Policy): Apply per-subscriber or per-service controls.
A. Define Class Maps
class-map match-any VOICE match dscp ef class-map match-any VIDEO match dscp af41 class-map match-any CRITICAL match dscp cs5 class-map match-any BULK match dscp cs1
B. Create Child Policy
policy-map CHILD-POLICY class VOICE priority level 1 class VIDEO bandwidth percent 25 class CRITICAL bandwidth percent 20 class BULK bandwidth percent 10 class class-default fair-queue
C. Create Parent Policy
policy-map PARENT-POLICY class class-default shape average 10G service-policy CHILD-POLICY
D. Apply to Interface
interface HundredGigE0/0/0/0 service-policy output PARENT-POLICY
E. Optional Grandchild Policy
policy-map GRANDCHILD-POLICY class SUBSCRIBER1 shape average 100M class SUBSCRIBER2 shape average 200M ! policy-map CHILD-POLICY class BULK service-policy GRANDCHILD-POLICY
F. Validate HQoS
show policy-map interface HundredGigE0/0/0/0 output show qos interface HundredGigE0/0/0/0 hierarchy show policy-map target parent all
The Bottom Line
Upgrading links doesn’t eliminate congestion — it just raises the ceiling. Cisco’s QoS toolset — from Catalyst’s MQC to Nexus VoQ to Cisco 8000 HQoS — gives you precise control of how traffic behaves under pressure. Hierarchical QoS is the crown jewel for WAN and core networks, ensuring fairness, priority, and determinism at scale.