It's worth pointing out that we're looking at more like 0.65c, through optical fibre. This would also be ignoring routing infrastructure/processing time and network prioritisation.
Fiber optic cables may not be able to penetrate the molten core of the Earth, but there is no law of physics that prevents information from traveling through the core of the Earth. Compare this to the speed of light, which is an absolute upper bound on how quickly information can propagate. There is simply no way, in this Universe, to do better than the straight-line distance through the center of the Earth.
There are no upper boundaries when it comes to latency, but at some point your browser/whatever will start to resend the information, thinking it's lost. Typically you should have around 20-30ms within the same country, depending on peering and such.
When you start to factor in things like the fact that light travels 1/3 as fast in a glass fiber as it does in vacuum, routing overhead, and the fact that ping is a round-trip measurement, 60-100ms is about normal.
15 milliseconds * 2 (ping = round trip time) * 1.52 (~index of refraction in fiber) = 45.6ms straight line. Add in routing dilays and less than optimal paths and your quickly hit 60-100 ms.
> The high latency with dsl usually comes hand in hand with interleaving enabled.
Interleaving doesn't add huge amounts of latency. To the LNS with the ISP I'm on, I usually see about 11-16ms to the first hop, maybe 5ms to a game server on their network a most. Without interleaving, I see about 5-7ms less on a good day.
Now, saying that, there are different levels of interleaving that can be applied on modern DSLAMs, but we're not really talking tens of milliseconds here.
It's also worth keeping in mind that your pings won't be truly representative of latency. Plenty of ISPs de-prioritise ICMP traffic.
My pings are really similar to my latency measured with different sources.
My isp (Qwest) uses interleaving and it adds quite a bit to my first hop (it usually is about 60ms) and from what I've seen those people actually living in a more populated area the interleaving latency added is about the same (>30ms first hop and usually closer to 60). So of course it depends on the ISP and qwest is notoriously bad for this but is my only option, and occurs with plenty of other ISPs with interleaving.
Well, ping is relative. In Norway we generally have around 6-8ms to Norwegian Internet Exchange, and typically end up at around 30-40ms to most parts of Europe. At some point distance will get you. Regular cobber is not actually slower, but the equipment connected to it is, and that is why you usually see at least 30ms+ just leaveing your ISP network.
For instance, if you're using WLAN you might end up doubling your response time. Something to keep in mind if you're a latency freak like me!
All that means is that Google's Stanford fiber is topologically close to where speedtest.net's SF server is hosted, which is apparently an ISP called Unwired. The fact that it didn't pick speedtest's Palo Alto server probably means that Google and Unwired share the same upstream fiber loop. Also interesting from the image is that speedtest identified the ISP as Tata Communications.
Edited for clarity.