Global Metro Bar Chart
Interconnect-quality ("pain") scores for every metro with two or more M-Lab servers — how differently the top client ISPs are served across the M-Lab servers in the metro. Lower is better; high values suggest some users may struggle to reach certain local content.
For a new user, this is the best place to start.
High scores suggest Anomalous Topology, Routing Policies, or Congested Interconnections in the mid-path between the ISPs in the region.
For the top N client (user-access) ISPs in each metro, it finds the worst-case performance difference between pairs of M-Lab sites serving that ISP. A healthy interconnection ecosystem produces small differences. MinRTT and throughput are scored separately: differences in minRTT indicate paths longer than necessary (routing or peering anomalies); differences in throughput suggest congestion or capacity problems between ISPs.
The two measures
- Spread (ratio) — the ratio between the highest and lowest per-site geometric-mean performance for an ISP. This has a strong intuitive meaning: 1.0 is ideal, 1.15 means the two differ by 15%. Under certain conditions spread can badly underestimate the difference between two sites.
- KSdistance × 10 — the maximum Kolmogorov–Smirnov distance between the per-site CDFs for an ISP (Wikipedia). More robust than the ratio but it has a poor intuitive meaning; scaled ×10 so 0 is best and 10 is worst (e.g. fully non-overlapping distributions).
Low scores on both are good.
Controls
| Control | Effect |
|---|---|
| Client ISPs | number of top-ranked ISPs (by test volume) to include. Too many pulls in tiny ISPs with poor connectivity and noisier results — balance is key. |
| Verbose | includes all single-server metros, which have no comparison and show a spread of 0. |
The date range is not an active selector — it shows the date range of the pre-computed cached data.
Navigation
Click a bar for a metro and a link appears below the chart; click that link to open the Regional Details dashboard with that metro preselected.