The docs say they use the time it takes to do a TCP open between their agents because they "believe that TCP Open times are more representative of Internet users' actual experience".
That is, they think pings could be treated in a different way but with a TCP connection establishment you are measuring what most internet users want to do over the link.
As for different readings from different directions you can get that with both pings and TCP open, it's just maybe more noticable and exacerbated with the longer and multiple-message nature of a TCP open.
This is awesome, I've been looking for something like this for a while.
It'd be great to expand it to the Tier2 providers in a undirected graph format, so you could see latencies from the peering points of various ISPs/datacenters. That way you could check on the real-world connectivity between places you interface with -- AWS, The Planet, Verizon in NY, Comcast in the Mid-Atlantic, etc.
6 comments
[ 3.9 ms ] story [ 28.1 ms ] threadThat is, they think pings could be treated in a different way but with a TCP connection establishment you are measuring what most internet users want to do over the link.
As for different readings from different directions you can get that with both pings and TCP open, it's just maybe more noticable and exacerbated with the longer and multiple-message nature of a TCP open.
It'd be great to expand it to the Tier2 providers in a undirected graph format, so you could see latencies from the peering points of various ISPs/datacenters. That way you could check on the real-world connectivity between places you interface with -- AWS, The Planet, Verizon in NY, Comcast in the Mid-Atlantic, etc.
http://www.keynote.com/products/web_performance/index.html