Overview & Thematic Scope
Quality of Service (QoS) shaping and hierarchical QoS (HQoS) are the control points where network engineers either win or lose their SLA commitments. This troubleshooting-focused FAQ answers the questions engineers actually ask when fine-tuning shapers, nesting HQoS levels, and chasing down drops, jitter, and underutilized circuits. Each answer leads with a direct verdict, then gives the configuration context you need to resolve the issue in production.

Frequently Asked Questions
- Q1: What is the difference between a QoS shaper and a policer, and which one should I deploy?
- Use a shaper on egress and a policer on ingress—shapers buffer excess traffic and smooth it out, while policers drop or remark excess traffic immediately. Shaping is correct for WAN sub-rate circuits where you must not exceed the provider’s committed rate; policing is correct at the edge to enforce contractual limits and protect downstream resources. Common error: applying a policer where a shaper is needed causes TCP global synchronization and severe throughput collapse.
- Q2: How do I configure hierarchical QoS (HQoS) so that per-subscriber shaping and per-class queuing work together?
- Build a three-level hierarchy: a root port/VLAN shaper, a per-subscriber (group) shaper, then per-class queues with scheduling inside each subscriber node. Apply the parent shaper first, attach the class-based child policy beneath it, and verify the parent’s shaping rate is not lower than the sum of child guarantees. In Cisco IOS-XE this is a hierarchical policy-map on the subscriber interface; in Junos it is a hierarchical scheduler with a logical interface shaper on top of a per-unit scheduler map.
- Q3: Why is my shaper showing drops or underutilization even though the circuit is not congested?
- The most common cause is a shaper rate set below the physical line rate without accounting for Layer 2 overhead, causing premature token exhaustion. Second most common is burst size tuned too small, which forces the shaper to drop or delay packets that would otherwise pass. Verify the configured committed information rate (CIR) against the actual carrier rate, and confirm the shaping burst equals at least the serialization time of one MTU at the shaped rate.
- Q4: What is the correct formula for calculating shaper burst size to avoid jitter and drops?
- Set the shaping burst to at least the serialization time of one full-size packet at the shaped rate—burst (bytes) ≈ shaped rate (bps) × MTU (bytes) × 8 / line rate. For example, a 100 Mbps shaper with a 1500-byte MTU needs roughly a 1.5 KB minimum burst, though most vendors recommend 2–4× that value for TCP-friendly behavior. Undersized bursts cause per-packet drops and jitter spikes; oversized bursts allow micro-bursts that violate the carrier’s policer.
- Q5: How do I validate that HQoS is actually shaping per subscriber rather than just at the port level?
- Check per-node statistics, not just interface counters—on Cisco use
show policy-map interface <intf>.service-instance <id>, on Junos useshow interfaces extensiveand hierarchical scheduler statistics. Confirm each subscriber node shows its own shaped rate and queue drop counters incrementing independently. If only the root node shows activity, the child policy is not attached to the subscriber context and HQoS is silently degraded to flat port shaping. - Q6: Which ASIC or platform limitations break hierarchical QoS on high-density line cards?
- Many merchant-silicon ASICs support only two or three HQoS levels and a fixed number of shaper nodes per port group, so deeply nested policies fail to program. Check the platform’s HQoS scale numbers—levels supported, shaper instances per ASIC, and queues per subscriber—before designing the hierarchy. Oversubscribed shaper tables are the leading cause of silent policy-programming failures and unexpected flat-rate behavior on 100G/400G line cards.
- Q7: How do I troubleshoot a configuration that applies cleanly but produces no shaping effect?
- Verify three things in order: the policy is bound to the correct direction (egress, not ingress), the interface is not running a conflicting legacy rate-limit, and the service policy is attached to the actual forwarding path (subinterface or service instance). Also confirm the QoS feature is enabled globally—some platforms require explicit
mls qosor equivalent before any policy takes effect. A clean commit with zero shaped bytes is almost always a bind-path mismatch. - Q8: What post-sales support and firmware guidance should I expect for fine-tuning HQoS in production?
- Require vendor TAC access to HQoS-specific escalation engineers, documented shaper-node scale limits per platform, and a firmware upgrade path that preserves policy syntax across releases. Ask for pre-upgrade validation of your hierarchy on the target release, since HQoS scheduler behavior and burst defaults have changed in major code trains. Retain a known-good configuration baseline and lab-validate every QoS-affecting upgrade before production rollout.
Leave a comment