Average
The mean round trip across every reply. This is the number people mean when they say ping, and the one to compare between regions.
Measure the round trip from the regions your visitors actually use. Latency is a distance problem, and one server cannot see it from one place.
No account, no email, no limit beyond a fair use rate limit. Results describe the path at the moment you ran it.
Reading the result
The mean round trip across every reply. This is the number people mean when they say ping, and the one to compare between regions.
How much the time varies, shown as mean deviation. Steady is good. Wild swings make calls and video choppy even when the average looks fine.
The share of packets that never came back. Even 1 percent hurts, because TCP reads loss as congestion and slows the whole connection down.
The address that answered and the network announcing it. If that is a CDN edge, you are measuring the edge rather than the origin.
What to expect
Light in fiber travels at roughly 8 microseconds per mile (5 microseconds per kilometer), and it must go there and back. That sets a floor no provider can beat. The table below is the physics, which is why a host in Sydney will always feel far away to a visitor in London.
| Between | Best case round trip | Feels like |
|---|---|---|
| Same city | Under 5 ms | Instant |
| Same country | 5 to 25 ms | Instant |
| Same continent | 20 to 60 ms | Immediate |
| Across an ocean | 120 to 180 ms | Noticeable |
| Antipodes | 250 to 300 ms | Sluggish |
These are floors, not targets. Anything meaningfully above them points at congestion, a long detour or a saturated link somewhere on the path.
Good to know
Plenty of firewalls drop ICMP deliberately, and a CDN will answer from whichever edge is nearest rather than from the machine you were thinking of. No reply means no reply, not necessarily an outage.
A server that answers in 4 ms from Cape Town can take 190 ms from Sydney. Run the test from each region you sell into, then decide whether one location is enough.
Next
A slow average points at distance. Loss and jitter point at a fault, and the quickest way to find whose it is, is to follow the path hop by hop.