LaunchTelephony

Performance Metrics

This page explains key telephony performance metrics and how to measure them.

Key Parameters

  • RTT (round‑trip time): How long it takes for your voice signal to go to the other person and back.
  • One‑way latency: The delay from when you speak to when the other person hears you.
  • PDV/jitter (packet delay variation): How much that delay changes over time — if it jumps around, voices can sound choppy.
  • Packet loss: How many bits of voice data get lost in transit — too much loss makes speech cut out.
  • MOS (Mean Opinion Score): A 1–5 quality score for how clear the call sounds overall.
  • RTT e2e (ms) — target < 100 ms for regional traffic.
  • One‑way latency (ms) — typically half of RTT; target < 50 ms for regional traffic.
  • PDV/jitter (ms) — keep p95 low and consistent to avoid buffer issues.
  • Packet loss (%) — keep p95 minimal; higher loss degrades speech quality.
  • MOS — target > 4.2 under healthy network conditions.
  • Call setup time (s) — SIP 100→200 end‑to‑end; monitor for delays.
  • Error rates — 4xx/5xx responses, RTP late/empty buffer events.

Tip
Sample across busy hours (e.g., 10:00–12:00 and 14:00–17:00 local) and compare weekdays.

Measurement Methodology

Use these steps to baseline your telephony infrastructure, validate improvements from regional deployment, or troubleshoot call quality issues. This methodology helps you collect statistically valid data to compare before/after scenarios.

  1. Network path

    • mtr -u -c 200 <media-endpoint> — UDP path with loss and jitter.
    • ping -c 200 <media-endpoint> — baseline RTT (may be ICMP‑filtered).

  2. Synthetic calls

    • Generate calls between your PBX and Synthflow; capture RTP with Wireshark and review RTP Stream stats (jitter, loss, skew).
    • Optionally use sipp to place load‑controlled calls and log response timings.

  3. Jitter/loss under load

    • If you can test off hours: iperf3 -u -b 2M -t 60 -c <test-host> with a corresponding server to observe UDP loss/jitter.
    • Compare public‑internet path vs Synthflow backbone path when available.

  4. MOS estimation

    • Use E‑Model‑based tools to derive R‑Factor and MOS from one‑way delay, jitter, and loss.

  5. Sampling window

    • Minimum 100 calls or 30 minutes per scenario (before/after, region A/B).
    • Record p50/p95/p99 for each metric.

Warning
Do not rely on a single call. Jitter and loss are bursty; use percentiles and a representative window.

Why PDV/jitter degrades speech

RTP packets carry audio frames assuming near‑constant arrival. When inter‑arrival times vary (high PDV/jitter), the jitter buffer may:

  • Stretch playback (robotic/slow speech) to avoid gaps,
  • Drop late packets (choppy audio), then
  • Catch up when bursts arrive (brief “squeal” or fast‑forward effect).

Keeping jitter low and consistent avoids buffer underflows/overflows and preserves natural cadence.