SIP Trunk Concurrency & Bandwidth Planner

1. Bridging Teletraffic Theory and Packet Networking

In legacy Time-Division Multiplexing (TDM) telephony, circuit sizing followed rigid physical constraints: a Primary Rate Interface (PRI) consisted of fixed 64 kbps DS0 timeslots—either 23 B-channels on a North American T1 span or 30 B-channels on a European E1 span. In modern Enterprise Unified Communications (UC) and Session Initiation Protocol (SIP - RFC 3261) trunking, physical timeslots are replaced by logical packet streams traversing shared IP Local Area Networks (LANs) and Wide Area Networks (WANs).

Dimensioning a carrier-grade SIP trunk requires a rigorous two-stage engineering methodology:

1. Teletraffic Concurrency Dimensioning (Erlang B Layer): Use Agner Krarup Erlang's classical M/M/m/m loss distribution to determine the minimum number of simultaneous, unblocked voice channels (m) needed to accommodate peak busy-hour calling volume at a contracted Grade of Service (GoS ≤ 1.0%).
2. IP Transport Packetization Sizing (Packet Network Layer): Translate those m concurrent sessions into Layer 3 IP and Layer 2 data link transmission rates based on audio codec sampling bitrates, packetization intervals (ptime), transport protocol headers, and control-plane signaling overheads.

Failure to execute both stages leads to catastrophic network degradation: under-sizing channels causes SIP 503 Service Unavailable call rejections at the PBX/SBC, while under-sizing IP WAN bandwidth induces packet loss, jitter buffer underruns, and degraded Mean Opinion Scores (MOS).

2. The Anatomy of a SIP Voice Packet: From Audio Sample to Wire Footprint

When human speech is digitized, analog acoustic pressure waves are sampled, quantized, and encoded into discrete binary frames. The length of time compressed into a single IP packet is determined by the packetization interval (ptime), which universally defaults to 20 milliseconds (0.020 seconds) in enterprise telephony to strike an optimal balance between acoustic latency and header encapsulation efficiency.

Consider standard uncompressed G.711 (PCMU/PCMA) operating at 64,000 bps:

For compression codecs such as G.729 (CS-ACELP at 8,000 bps), the payload shrinks dramatically:

To transmit this audio across an IP network, protocol headers are prepended at each layer of the OSI reference model:

• Real-Time Transport Protocol (RTP - RFC 3550): 12 Bytes (sequence numbers, timestamps, SSRC synchronization identifiers).
• Secure RTP (SRTP - RFC 3711): Adds 4 Bytes for authentication tags (HMAC-SHA1) and optional Master Key Identifier (MKI).
• User Datagram Protocol (UDP - RFC 768): 8 Bytes (source and destination port numbers, length, checksum).
• Internet Protocol (IPv4): 20 Bytes (or 40 Bytes for IPv6).
• Layer 2 Data Link (Ethernet): 14 Bytes MAC destination/source + 4 Bytes CRC/FCS = 18 Bytes (plus 4 Bytes for 802.1Q VLAN or 8 Bytes for MPLS labels).
• Physical Layer 1 Framing (Ethernet): 7 Bytes preamble + 1 Byte Start Frame Delimiter (SFD) + 12 Bytes Inter-Packet Gap (IPG) = 20 Bytes.

On a standard IPv4 Ethernet network, protocol headers contribute 78 Bytes of overhead per packet. For G.729, the header overhead (78 Bytes) is nearly four times larger than the voice audio payload itself (20 Bytes), highlighting the profound impact of encapsulation math on WAN provisioning.

3. SIP Signaling Overhead & Transaction Rate Budgeting

While RTP media flows continuously throughout the duration of a call, Session Initiation Protocol (SIP - RFC 3261) signaling messages are transmitted during session initiation, mid-call renegotiation, and session termination.

A standard call setup sequence consists of an initial INVITE message containing Session Description Protocol (SDP) codec negotiation offers, followed by 100 Trying, 180 Ringing, 200 OK, and final ACK handshakes. Call teardown requires a two-message BYE / 200 OK exchange.

The peak Call Setup Rate (Calls Per Second / CPS) during the busy hour is calculated directly from Busy Hour Call Attempts (BHCA):

Because individual SIP signaling packets range from 500 to 1,500 bytes, steady-state signaling rarely exceeds 15 to 50 kbps even on trunks supporting thousands of calls. However, network architects must reserve a 5% to 10% bandwidth margin above raw media to guarantee that signaling packets (tagged with DSCP CS3 or AF31) do not experience buffer drops during bursty call setup storms or carrier trunk failover renegotiations.

4. Session Border Controller (SBC) & DSP Transcoding Dimensioning

A Session Border Controller (SBC)—such as an AudioCodes Mediant, Ribbon SBC SWe, Cisco CUBE, or Oracle Acme Packet—is deployed at the demarcation boundary between the enterprise private IP network and the public carrier SIP trunk. The SBC enforces topology hiding, NAT traversal, firewall pinholing, Denial-of-Service (DoS) rate limiting, and protocol normalization.

The Transcoding Bottleneck: When internal IP-PBX phones negotiate high-definition wideband audio (G.722 or Opus) while the external telecommunications carrier only supports standard PSTN G.711, the SBC must perform real-time Digital Signal Processing (DSP) transcoding. Transcoding requires decoding the inbound RTP packet into linear PCM samples and re-encoding it into the target format.

Because DSP transcoding consumes significant CPU cycles and dedicated hardware resources, enabling transcoding for 50% or 100% of calls typically reduces an SBC's maximum rated session concurrency by 50% to 70%. Enterprise architects must dimension both logical SIP channel licenses and physical DSP hardware modules to avoid mid-call clipping and audio degradation.