Packet Loss Impact Calculator – TCP Throughput, UDP & VoIP Quality
The Packet Loss Impact Calculator quantifies how dropped packets degrade real-world network performance. Enter your link speed, packet loss rate, and round-trip time to instantly see the effective TCP throughput, UDP throughput, retransmission overhead, VoIP MOS score, and per-application quality ratings. The tool uses the Mathis TCP throughput model and the ITU-T E-Model for accurate, standards-based estimates.
Why Packet Loss Matters More Than You Think
Even a small amount of packet loss causes disproportionate harm to TCP-based applications. TCP's congestion-control algorithm (AIMD — Additive Increase Multiplicative Decrease) interprets every lost packet as a sign of network congestion and halves the transmission window. At 1% packet loss with a 40 ms RTT, a 1 Gbps link delivers only around 13 Mbps of usable TCP throughput — a 98.7% reduction. This non-linear sensitivity is why packet loss is one of the most impactful network quality metrics.
TCP Throughput: The Mathis Formula
The calculator uses the widely cited Mathis formula to estimate maximum TCP throughput:
T = (MSS / RTT) × (C / √p)
Where:
- MSS — Maximum Segment Size (default: 1460 bytes for standard Ethernet with a 1500-byte MTU)
- RTT — Round-Trip Time in seconds
- p — packet loss probability (0–1)
- C — a constant ≈ 1.22 (√(3/2), standard Mathis value)
For multiple parallel TCP flows, the aggregate throughput scales linearly with the number of flows, each probing independently. The result is clamped to the nominal link speed.
UDP Throughput and Retransmission Overhead
UDP has no retransmission mechanism. The effective UDP throughput is simply nominal × (1 − p), making it far less sensitive to loss than TCP — which is why real-time media (VoIP, video streaming, gaming) commonly uses UDP despite receiving no delivery guarantees.
The retransmission overhead shows how much of your link capacity TCP wastes re-sending dropped packets: overhead = nominal × p / (1 − p). At 5% loss, roughly 5.26% of your bandwidth is consumed by retransmissions alone, on top of the congestion-window throttling.
VoIP Quality: MOS Score via the ITU-T E-Model
The Mean Opinion Score (MOS) is an ITU-T standard for rating voice call quality on a 1–5 scale. The calculator uses a simplified E-Model estimate:
R = 93.2 − (loss% × 2.5)MOS = 1 + 0.035R + R×(R−60)×(100−R)×7×10⁻⁶
Scores are clamped to the range 1.0–4.5. Business-grade VoIP typically requires a minimum MOS of 4.0. The tool also reports the maximum tolerable packet loss percentage that keeps VoIP at the "Good" level — typically around 0.5%.
Application Quality Thresholds
Different applications tolerate packet loss differently. The dashboard rates four traffic types against standard industry thresholds:
- File Transfer (TCP) — Good below 1%, Poor above 2.5%. TCP eventually delivers all data, but throughput collapses severely at high loss rates.
- VoIP — Good below 1%, Poor above 3%. Listeners hear clipping, dropouts, and robotic artefacts above 1%.
- Video Streaming — Good below 0.5%, Poor above 2%. Buffering, pixelation, or frozen frames appear above the threshold.
- Online Gaming — Good below 0.5%, Poor above 1%. Rubber-banding, teleporting characters, and disconnections are common at even moderate loss rates.
Using the Sensitivity Table
The Throughput vs Packet Loss Sensitivity table shows TCP and UDP throughput at standard loss percentages (0%–10%), computed with your current bandwidth and RTT settings. This lets you quickly gauge the full impact curve — for example, seeing how a link that performs adequately at 0.5% loss becomes critically impaired at 2%.
Practical Guidance for Network Engineers
When diagnosing network problems, target less than 0.1% packet loss for all production applications. Investigate loss above 0.5% immediately. Common causes include congested uplinks, faulty cables or transceivers, misconfigured QoS policies, wireless interference, and buffer bloat. Tools like ping, mtr, and iperf3 can help localise which hop in the path is dropping packets.
Use the Number of Flows advanced option to model multi-stream downloads (e.g., web browsers opening 6–8 parallel connections, or BitTorrent-style transfers). Parallel flows partially recover aggregate throughput because each independently probes the congestion window.