Latency by country
Pick a country to see every service its probes measure.
Services measured from here
Latency over time
One line per check type — the median across every service of that type measured from here, log scale. The faint band is the spread between the fastest and slowest service in the group. DNS sits near the floor and TLS near the ceiling; that separation is the point. Individual services have their own sparkline in the table above and their own row in the distribution below.
Distribution
How each service behaves from here across the whole range.
- One row per service, as measured from this country's probes. Every measurement in this range falls into one of fourteen buckets, chosen by how long it took.
- The numbers across the top are the bucket edges — milliseconds up to 50, then seconds:
.1sis 100 ms,1sis 1000 ms. The spacing is logarithmic, so each step right is a much slower answer than the last.∞holds anything over 10 seconds; its real value is not kept. - Bar height is scaled inside its own row. The tallest bucket in a row is always full height, so bars compare buckets within one probe — never one probe against another. The shape is what to read: a single solid block is a steady connection, two separated humps mean two different paths or two different servers answering.
- p95 on the right — 95% of measurements came back at or under this. It is derived from the summed histogram across the whole range, which is why it stays correct at 7d and 30d. Percentiles cannot be averaged; the counts they come from can be added.
- p95 is interpolated inside the bucket it lands in, so read it as a magnitude, not an exact figure.
- This panel is built from hourly rollups, so it stays empty on the 1h range.