AWS data center latencies, visualized

Original Article ↗ Hacker News Discussion ↗

Visualization & UX

  • Many like the rotating 3D globe aesthetically but find it hard to read exact latencies and see all regions at once.
  • Suggestions:
    • Add a hint like “click to select a region” and auto-select a default or nearest region.
    • Offer a 2D map/projection (e.g., Winkel Tripel, azimuthal equidistant) or toggle between 3D/2D.
    • Add or expose a table view alongside the visualization.
    • Use a continuous color ramp instead of coarse latency buckets.

Data & Methodology

  • Points shown are AWS regions, not individual data centers; each region contains multiple AZs and facilities with different intra-region characteristics.
  • Latency data is scraped from CloudPing, which:
    • Uses TCP connections to AWS services (not ICMP ping).
    • Hasn’t been updated in years, so newer regions are missing.
  • Some note AWS also exposes latency baselines via Network Manager, but reliability of provider dashboards is debated.

Latency Physics & Infrastructure

  • Distance is the dominant factor, but:
    • Speed of light in fiber is ~2/3 of c, plus extra delay from non-straight paths, switching, and routing.
    • Rules of thumb like “distance(km) / ~150 ≈ ms” or ~125 miles/ms appear to fit many cases.
  • Undersea cable layout strongly shapes latency; absence of direct cables (e.g., Ireland ↔ continent, South America/Africa) forces longer routes.
  • Some propose showing “overhead vs theoretical minimum” per link.

Coverage & Missing Regions

  • Multiple AWS regions (e.g., Spain, Melbourne, Israel, Hyderabad, Canada West) are absent due to CloudPing’s dataset.
  • Users observe sparse coverage in Africa and South America and discuss likely reasons: smaller markets, power/heat challenges, political risk, and limited subsea capacity.

Use Cases, Alternatives & Ideas

  • People have used similar measurements for:
    • Region selection, HA/DR design, and performance tuning (e.g., impact of cross-US region hops).
    • Trading and latency-sensitive workloads, sometimes using microwave or specialized fiber.
  • Requests for:
    • Versions for Azure/GCP and inter-cloud latency.
    • A mode showing user-to-region latency, history charts, and inferred “likely fiber” links.

Accessibility & Feedback Culture

  • Red/green color choices are hard for color-blind users and grayscale displays; suggestions include:
    • Color-blind-friendly palettes.
    • Line patterns (dashed/dotted) and an optional “color-blind mode.”
  • Some debate whether such critiques are discouraging or valuable; others argue accessibility feedback is important even for hobby projects.