Overview

Configuring QoS to prioritize VoIP traffic on an ONT requires classifying SIP/RTP packets, marking them with high-priority DSCP or 802.1p values, and mapping them to a strict-priority egress queue. This FAQ answers the most common pre-sales and post-sales questions from network architects and support engineers deploying voice services on enterprise ONTs. The guidance below applies to generic carrier-grade ONTs with a Layer 2/3 QoS engine, covering CLI configuration, VLAN tagging, DSCP trust boundaries, buffer tuning, and failover verification.

Frequently Asked Questions

Q1: How do I configure QoS to prioritize VoIP traffic on an ONT using the CLI?

You configure QoS for VoIP on an ONT by creating a classifier that matches SIP (UDP 5060) and RTP (UDP 16384-32767), marking those packets with DSCP EF (46) or 802.1p priority 5, and assigning them to a strict-priority egress queue. A typical CLI sequence is: qos classify match udp 5060 mark dscp ef, qos classify match udp 16384-32767 mark dscp ef, qos queue 7 strict-priority, then qos map dscp ef queue 7. Apply the policy to the WAN uplink and the voice VLAN, then verify with show qos statistics that EF packets increment on queue 7 and not on best-effort queues.

Q2: What DSCP and 802.1p values should I use to prioritize VoIP on an ONT?

The recommended markings are DSCP EF (46) for RTP media and DSCP CS3 (24) or AF31 (26) for SIP signaling, with 802.1p priority 5 (or 6 for call control) on the voice VLAN tag. DSCP EF maps to the highest strict-priority queue and guarantees low latency, jitter, and packet loss for G.711 and G.729 streams. If your ONT only supports Layer 2 QoS, use 802.1p priority 5 on the voice VLAN and set the uplink port to trust 802.1p. Always align these values with your carrier's SLA and the upstream router's QoS policy to avoid re-marking or dropping.

Q3: Which ONT QoS mode is best for VoIP: strict priority, WFQ, or shaping?

Strict priority (SP) is best for VoIP because it services the EF queue before all other queues, minimizing jitter and latency. Weighted Fair Queuing (WFQ) is suitable only when voice traffic is a small percentage of total bandwidth and you need to prevent starvation of data queues; a hybrid SP+WFQ scheduler with SP reserved for EF is the common enterprise choice. Traffic shaping should be applied only to best-effort and bulk queues, never to the voice queue, because shaping adds delay. Configure a small guaranteed bandwidth reservation (for example 10-15% of uplink) for the voice queue plus SP scheduling for deterministic voice quality.

Q4: How much bandwidth and buffer should I reserve for VoIP on an ONT?

Reserve approximately 100 kbps per concurrent call for G.711 (plus 10-15 kbps for SIP signaling) and 30-40 kbps per call for G.729, then add 20-30% headroom for concurrent calls and retransmissions. For a typical enterprise ONT supporting 100 concurrent calls, a 15-20 Mbps voice reservation with a 64-128 KB dedicated egress buffer per queue is sufficient. Keep the voice queue buffer shallow (10-20 ms of line-rate traffic) to avoid bufferbloat; deep buffers increase jitter even when latency looks acceptable. Monitor with show qos queue-stats for drops and adjust the reservation based on peak concurrent call counts.

Q5: Does the ONT need to trust DSCP from the IP phone or re-mark traffic?

The ONT should trust DSCP from the IP phone when the phone is managed and marks EF correctly; otherwise the ONT must re-mark untrusted traffic at the ingress port. Best practice is to configure the access port as qos trust dscp for the voice VLAN and qos trust none or qos mark dscp 0 for data VLANs, then police the voice VLAN to prevent abuse (for example, rate-limit EF to the provisioned call capacity). This trust boundary prevents a compromised endpoint from flooding the strict-priority queue and starving business-critical data. Verify the trust configuration with show qos interface and packet captures on the uplink.

Q6: How do I troubleshoot VoIP quality issues after enabling QoS on the ONT?

Start by confirming the voice packets are actually marked and queued correctly: run show qos statistics interface wan and check that EF counters increment on the strict-priority queue. If counters stay at zero, the classifier is not matching (wrong UDP range, wrong VLAN, or DSCP trust disabled); if counters increment but voice quality is poor, check upstream congestion, jitter, and whether the carrier is re-marking DSCP to 0. Use show qos queue-drops to detect voice queue drops, verify the uplink is not oversubscribed, and confirm the voice VLAN is not being shaped by a parent policy. Finally, test with a packet capture and a MOS score tool to isolate ONT QoS from WAN or call-server issues.

Q7: What happens to VoIP QoS if the ONT loses power or fails over to a redundant PSU?

QoS configuration is stored in non-volatile memory and is restored automatically after a power failover or PSU switchover, so voice prioritization resumes within the ONT boot time (typically 30-90 seconds). During the reboot window, voice calls drop and the uplink is unavailable; to avoid this, deploy dual redundant AC/DC PSUs on the same ONT and connect them to separate power feeds. After failover, verify with show qos config that the classifier and queue mapping are intact, and confirm the strict-priority queue counters increment once calls resume. For zero-downtime requirements, use a redundant ONT pair with VRRP or MC-LAG rather than relying on a single chassis.

Q8: Can I prioritize VoIP on an ONT that only supports Layer 2 QoS?

Yes. On a Layer 2-only ONT, prioritize VoIP using 802.1p priority 5 or 6 on the voice VLAN tag and map that priority to a strict-priority egress queue. Configure the access port to trust 802.1p for the voice VLAN, set the uplink to preserve the tag, and reserve bandwidth for the voice queue. The main limitation is that Layer 2 QoS cannot inspect DSCP inside the IP header, so if the upstream carrier or router re-marks or strips 802.1p, voice priority is lost; in that case, terminate voice on a Layer 3 ONT or upgrade to a model with DSCP classification. Always verify end-to-end marking with a capture at the uplink and at the far end of the WAN.