M-Lab Network Annotations: Geolocation, ASNs, and What They Mean

Every M-Lab measurement is enriched with metadata about the client’s network and location — annotations. Understanding what these annotations represent (and where they fall short) is essential for doing rigorous analysis with M-Lab data.

How Annotations Are Added#

Annotations are derived from the client’s IP address. For ndt7 tests (since 2020-03), M-Lab servers add them while the test runs. For older data, they were added during processing, using the database versions from around the test date. The IP address is looked up in several databases to add:

• Geographic location (country, region, city, lat/lon) from MaxMind GeoLite2
• Autonomous System Number (ASN) from RouteViews and CAIDA prefix-to-AS data
• AS name from IPinfo’s ASN database

This means annotations reflect these databases as they were at about the time of the test.

Geographic Annotations#

Fields Available#

client.Geo.CountryCode        — ISO 3166-1 alpha-2 (e.g., "US", "DE", "BR")
client.Geo.CountryName        — Human-readable country name
client.Geo.Subdivision1ISOCode — ISO 3166-2 subdivision code, without the country prefix (e.g., "CA" for California)
client.Geo.Subdivision1Name   — Human-readable subdivision name
client.Geo.City               — City name (MaxMind estimate)
client.Geo.Latitude           — Latitude (see accuracy notes below)
client.Geo.Longitude          — Longitude (see accuracy notes below)
client.Geo.PostalCode         — Postal code (not available for all countries, unreliable)
client.Geo.AccuracyRadiusKm   — Estimated accuracy radius in km

Accuracy and Limitations#

Country-level IP geolocation is generally suitable for coarse aggregation, especially in major markets, but it should still be treated as an inferred database annotation rather than ground truth.

City-level geolocation is much less reliable. Published accuracy numbers can be misleading because they depend on the evaluation dataset, the definition of “city-level,” and the type of IP address being geolocated.

Coordinates are representative locations, not exact user or server locations. In M-Lab data, latitude and longitude should not be interpreted as device-level coordinates for clients or as exact physical coordinates for M-Lab servers. Many measurements attributed to the same city may share the same coordinate, often corresponding to a city centroid or another representative database location. This means that both the client-side and server-side coordinates are useful for coarse geographic aggregation, but not for neighborhood-, building-, or infrastructure-level analysis.

Do not use latitude/longitude for fine-grained spatial analysis Even when coordinates appear precise, the underlying geolocation may only be accurate at city scale. For sub-national analysis, prefer client.Geo.Subdivision1ISOCode over point coordinates when state/province-level aggregation is sufficient.

Rural, mobile, and less-populated areas require particular caution. Geolocation quality varies substantially by region, ISP, access technology, and address block. Errors are not uniform, so a single global accuracy number can obscure the cases where geolocation is least reliable. Recent measurement work shows that errors can vary by network type and geography: fixed-network IPs may have relatively small median errors, while mobile and Global South prefixes can have much larger errors and higher failure rates (see, for example, “Lost in the Prefix: Revisiting IP Geolocation Accuracy Across Networks and Geographies”).

Improving Spatial Precision#

If your analysis requires finer geographic resolution:

1. Filter by accuracy radius — use client.Geo.AccuracyRadiusKm, but treat it as a confidence signal rather than a guarantee.
2. Use M-Lab server-selection metadata to identify likely errors — M-Lab uses one geolocation system at test time to select a nearby server and another system to annotate the public dataset. As described in M-Lab blog post Improving M-Lab Geolocation, researchers can compare the server that actually served the test with the nearest server implied by the published client geolocation. Large inconsistencies between the two can flag potentially incorrect client geolocation.
3. Avoid overinterpreting city labels — city-level labels are useful for coarse aggregation, not for neighborhood-, block-, or infrastructure-level claims.
4. Aggregate spatially — use larger geographic cells or administrative regions rather than individual latitude/longitude points.
5. Use Subdivision1ISOCode for sub-national analysis — ISO 3166-2 region codes are generally more appropriate for state/province-level analysis than city centroids.
6. Validate with supplementary sources when precision matters — for rural, mobile, infrastructure-specific, or policy-sensitive analyses, M-Lab’s built-in geolocation should be supplemented with additional geolocation sources or ground-truth validation.

M-Lab’s built-in geolocation is appropriate for coarse spatial summaries, but it is not suitable for block-level, household-level, or infrastructure-specific analysis without additional validation.

Region Codes#

M-Lab uses ISO 3166-2 codes for subdivisions such as states and provinces:

-- US state-level analysis
SELECT
  client.Geo.Subdivision1Name AS state,
  COUNT(*) AS tests,
  ROUND(AVG(a.MeanThroughputMbps), 2) AS avg_mbps
FROM `measurement-lab.ndt.unified_downloads`
WHERE client.Geo.CountryCode = 'US'
  AND date BETWEEN '2024-01-01' AND '2024-01-02'
GROUP BY state
ORDER BY tests DESC;

Network Annotations (ASN)#

Fields Available#

client.Network.ASNumber     — Autonomous System Number (integer)
client.Network.ASName       — AS name from IPinfo (e.g., "Comcast Cable")
client.Network.CIDR         — IP prefix the client address belongs to (e.g., "73.0.0.0/8")

What ASN Annotations Mean#

ASN annotations identify the Autonomous System that appears to originate the prefix containing the client IP address in the routing data used by M-Lab’s annotation pipeline. In practice, this makes ASNumber a useful way to group measurements by network.

IP-to-AS mappings are derived from routing data and can be affected by route visibility, MOAS prefixes, third-party address use, resellers, VPNs, and stale or coarse-grained prefix mappings. They identify the AS announcing the client prefix, not necessarily the user’s retail ISP. ASN annotations should thus be treated as inferred metadata rather than perfect ground truth.

The accompanying ASName field provides a human-readable name for that ASN. This name is useful for display and interpretation, but it should not be treated as a stable identifier. For ISP comparisons, prefer ASNumber over ASName. AS names can change due to mergers and rebranding, whereas ASNs are more stable identifiers. Even so, some organizations operate multiple ASNs, and some ASNs contain multiple brands or customer populations.

-- Top ISPs by test volume in a country
SELECT
  client.Network.ASNumber AS asn,
  MAX(client.Network.ASName) AS isp_name,
  COUNT(*) AS test_count,
  ROUND(APPROX_QUANTILES(a.MeanThroughputMbps, 100)[OFFSET(50)], 2) AS median_mbps
FROM `measurement-lab.ndt.unified_downloads`
WHERE client.Geo.CountryCode = 'BR'
  AND date BETWEEN '2024-01-01' AND '2024-01-01'
GROUP BY asn
HAVING test_count > 1000
ORDER BY test_count DESC

Corporate vs. Residential Networks#

A single ASN may include both corporate and residential customers. Comcast’s AS7922, for example, includes business customers who may have very different service tiers than residential subscribers. For research on residential broadband specifically, consider filtering by test source (app/platform embedding) if that metadata is available.

Geolocation Improvements Over Time#

M-Lab has invested in improving geolocation accuracy. Key milestones:

• 2017-09 — Moved from MaxMind GeoLite Legacy to GeoLite2, which uses ISO 3166-2 subdivision codes
• 2020-03 — ndt7 tests annotated at test time by the uuid-annotator
• 2022-03 — Unified views standardized on the ISO 3166-2 subdivision fields (Subdivision1ISOCode)

Matching M-Lab Data to Other Datasets#

When joining M-Lab data with external datasets (census, FCC broadband maps, etc.):