CAPACITY PLANNING / 5 MIN READ

Size the Voice Path, Not Just the Codec.

A codec describes audio payload. The network carries packets, with headers and other traffic competing for space.

Read the Result in Each Direction

The model assumes one continuous audio stream in each direction for every concurrent call. Compare its upload figure with the upstream capacity allocated to voice, and its download figure with the downstream allocation. Adding them together does not tell you the upload requirement.

Conference bridges, recording, media relays and additional call legs can change which streams cross a particular link. Count media on the bottleneck you are planning—not every registered extension in the organization.

Follow One Packet

The base model adds a 12-byte RTP header and an 8-byte UDP header to the base IP header: 20 bytes for IPv4 or 40 for IPv6. At 20 ms per packet, each stream sends 50 packets per second. Extensions and additional encapsulation are not included automatically.

Worked example: 10 G.711 calls over IPv4

64 kbps audio + (40 header bytes × 8 ÷ 20 ms)
= 80 kbps per call, per direction

10 × 80 × 1.25 ÷ 1,000
= 1.000 Mbps upload and 1.000 Mbps download

The 25% added margin multiplies the modeled load by 1.25. It leaves 20% of that resulting capacity unused by the modeled load—not 25%.

Choose an Accounting Boundary

The optional Ethernet setting adds 18 bytes for the MAC header and frame check sequence; one VLAN tag adds another 4. For G.711, IPv4 and 20 ms packets, the 18-byte model gives 87.2 kbps per call per direction before margin. These settings do not include preamble, inter-frame gap, Wi-Fi contention or retransmissions.

Use additional bytes only for overhead you have identified. An IP-layer interface counter and physical wire utilization need not agree. Margin is not a substitute for understanding that difference.

Packet Interval Is a Trade-off

Shorter intervals send more headers each second. Longer intervals reduce header cost but accumulate more audio before sending and put more audio in a single lost packet. The calculator offers 10, 20 and 40 ms as planning choices; endpoint support and the negotiated configuration still determine the real interval.

G.722 here uses its 64 kbps mode. Opus entries are target rates for an adaptive codec, not fixed rates discovered from your session. Silence suppression, variable bitrate and forward error correction can change actual traffic.

Before You Use the Number

Confirm busy-hour concurrent calls and the media path.
Match codec, IP version and packet interval to the actual deployment.
Budget separately for signaling, RTCP, video, redundancy, unmodeled security/tunnels and competing applications.
Check upstream and downstream capacity, queue behavior, MTU and call admission controls with the administrator.
Validate the plan under a representative, authorized workload.

Return to the Bandwidth Calculator →

Further reading: Cisco voice bandwidth calculations · RTP and RTCP — RFC 3550 · Opus over RTP — RFC 7587.

Updated September 6, 2026. Corrections and field feedback: [email protected]. Please leave customer identifiers and sensitive call records out of your first message.