Beyond Speed: Understanding Internet Quality Metrics

“How fast is your internet?” is the wrong question. Speed is one dimension of a multi-dimensional problem. M-Lab and the broader internet measurement community have been developing richer frameworks for understanding what makes an internet connection work well — or not.

Why Speed Isn’t Enough

Download throughput matters for large file transfers and video streaming, but many everyday internet experiences are more sensitive to other metrics:

Use case	Most important metric
Video calls (Zoom, Teams)	Latency + jitter + packet loss
Gaming	Latency + jitter
Web browsing	Latency (time to first byte)
VoIP	Packet loss + jitter
File download	Throughput
4K streaming	Throughput + buffering

A 1 Gbps connection with 200 ms latency or 2% packet loss will perform worse for video calls than a 50 Mbps connection with 10 ms latency and 0% loss.

Key Quality Metrics

Latency (Round-Trip Time)

Minimum RTT is the best-case round-trip time between your device and the measurement server — essentially the speed-of-light delay plus processing. It’s a property of the physical path, not congestion.

Working latency (sometimes called “loaded latency”) is the round-trip time during a data transfer. When a connection is busy sending or receiving data, latency often spikes dramatically — this is the real-world experience of lag during uploads or video calls.

In M-Lab NDT7 data:

-- NDT7 Data Query
SELECT
  a.MinRTT,         -- best-case (unloaded) latency in ms
  a.MeanThroughputMbps
FROM `measurement-lab.ndt.ndt7`
WHERE date = '2024-06-01'

Packet Loss

a.LossRate in NDT7 data measures the fraction of packets retransmitted during the test, as a proxy for packet loss. Values above 1–2% indicate significant congestion or link-layer errors.

Jitter

Jitter is variation in packet delay — even if average latency is low, high jitter makes real-time applications like video calls unreliable. NDT7 doesn’t directly measure jitter; MSAK measurements include jitter estimates.

The Internet Quality Barometer (IQB)

In 2024–2025, M-Lab began developing the Internet Quality Barometer (IQB) — a composite metric that combines throughput, latency under load, and packet loss into a single quality score. The IQB is designed to:

• Capture “working quality” rather than peak speed
• Be comparable across different measurement methodologies
• Be meaningful for end users and policymakers, not just engineers
• Account for different use cases (video, gaming, browsing)

The IQB builds on the “responsiveness” concept from Apple’s RPM test and BITAG’s working latency research. M-Lab is working to publish IQB methodology and include IQB scores in BigQuery.

Rate Limits and Test Integrity

M-Lab enforces a limit of 40 tests per day per IP address. This prevents bulk automated testing that would distort population-level statistics. If you need to run frequent tests for monitoring purposes, consider:

• Running a private NDT7 server (see run-a-private-ndt-server)
• Using MSAK which has separate rate limits
• Aggregating your analysis over longer time windows

See the Test Rate Limits FAQ for details.

Geolocation and What It Means for Quality Analysis

M-Lab uses MaxMind GeoLite2 to annotate measurements with geographic information. Important caveats:

• Geolocation is at city level at best — coordinates represent the city centroid, not the user’s location
• Privacy is the reason for coarse precision — M-Lab data is public, so fine-grained location data would expose users
• ASN annotations are more reliable than geolocation for identifying the network a user is on
• ISP-level analysis (using client.Network.ASNumber) is generally more reliable than sub-city geographic analysis

See Geolocation Limitations in M-Lab Data for how to handle these constraints in analysis.

What “Good” Looks Like

Rough benchmarks for residential broadband:

Metric	Adequate	Good	Excellent
Download	25 Mbps	100 Mbps	1 Gbps
Upload	3 Mbps	20 Mbps	500 Mbps
Latency (idle)	<50 ms	<20 ms	<10 ms
Packet loss	<1%	<0.1%	~0%

These benchmarks vary by use case. The FCC uses 25/3 Mbps as a minimum definition; the BEAD program uses 100/20 Mbps as the funded build-out target.

---