VoIP QoS Configuration

1. Why VoIP needs QoS

VoIP is extremely sensitive to network conditions. RTP packets must arrive within tight timing windows — jitter above 40ms causes audio degradation, and packet loss above 1% causes choppy calls. On a shared network without QoS, a single large file transfer or video stream can starve voice traffic and destroy call quality.

QoS solves this by marking voice packets with a high-priority DSCP value and configuring routers to serve those packets first, before data traffic. Even on a congested network, voice gets through.

2. DSCP EF marking for VoIP

DSCP (Differentiated Services Code Point) is a 6-bit field in the IP header that marks packet priority. For VoIP:

Mark RTP traffic on Linux/Asterisk

Mark in Asterisk directly

3. Priority queuing strategies

LLQ (Low Latency Queuing) — recommended for VoIP

LLQ gives a strict priority queue for EF traffic. Voice packets are always served before data. The voice queue has a bandwidth guarantee and a maximum to prevent it from starving data entirely.

tc/HTB on Linux

4. VLAN separation for VoIP

Putting voice on a dedicated VLAN is the most reliable QoS approach — it physically separates voice traffic from data at the switch level, not just at the router.

• Create a dedicated Voice VLAN (e.g., VLAN 10) on your switches
• Configure IP phones to use the Voice VLAN (most phones auto-detect via LLDP/CDP)
• Set 802.1p CoS=5 on the Voice VLAN for layer 2 priority
• Configure inter-VLAN routing with QoS policies on the router/L3 switch
• Ensure PBX has a NIC on the Voice VLAN or a trunk port with both VLANs

5. WAN QoS for VoIP

WAN links are where QoS matters most — they're typically the bottleneck. Key principles:

• Bandwidth allocation — reserve 20-30% of WAN bandwidth for voice, never more than 75% of total
• Each G.711 call uses ~87 kbps (including IP/UDP/RTP overhead). G.729 uses ~32 kbps.
• Upload is often the limiting factor — asymmetric internet links have less upload bandwidth
• Apply QoS before the bottleneck — mark and queue on the LAN side of the WAN router, not after

For SD-WAN deployments, most platforms have voice-aware QoS policies built in. Configure the voice application category to use EF marking and strict priority delivery.

6. Verifying QoS is working

The key test: run a speed test while on a VoIP call. With QoS properly configured, the call should remain clear even at 100% bandwidth utilization. Without QoS, the call will degrade or drop.

Frequently asked questions

What DSCP value should I use for VoIP?

Use DSCP EF (Expedited Forwarding, value 46) for RTP media packets and DSCP CS3 (value 24) or AF31 (value 26) for SIP signaling. DSCP EF ensures voice packets get strict priority treatment in router queues. In Asterisk set tos_audio=ef in asterisk.conf. In Linux use iptables mangle to mark UDP packets in the RTP port range with DSCP 46.

How much bandwidth does a VoIP call use?

A G.711 (PCMU/PCMA) VoIP call uses approximately 87 kbps per call including IP, UDP, and RTP headers. G.729 uses approximately 32 kbps per call. Opus at full quality uses around 510 kbps but can be configured as low as 6 kbps. For QoS planning, calculate concurrent calls times per-call bandwidth and reserve that plus 20% headroom on your WAN link.

What is the difference between DSCP and 802.1p QoS?

DSCP operates at Layer 3 (IP header) and is preserved end-to-end across routed networks. 802.1p CoS operates at Layer 2 (Ethernet frame) and only applies within a switched VLAN segment — it is stripped at router boundaries. For VoIP, use both: 802.1p CoS=5 for within the LAN/VLAN, and DSCP EF=46 for WAN and routed segments.