4 ms·

Latency and peering are huge. I'm amazed at how poor internet latency constantly is. Assuming fiber optics have a speed of light around 0.6c, one would expect t

by aftbit 2y ago

Latency and peering are huge. I'm amazed at how poor internet latency constantly is. Assuming fiber optics have a speed of light around 0.6c, one would expect that Cincinnati to Chicago (~450 km) is around 2.5 light-milliseconds one way, but actual pings take me at least 10ms, usually more like 20. Worse, my home ISP has pretty bad peerings, so for example, traffic from my home to a data center about thirty minutes drive away (well under 50km as crow flies) ends up going via Ashburn and Washington, which results in RTT in excess of 30ms. There is so much more room for improvement in latency. Even pinging my friend's house (under 10km away) on the SAME ISP (with a total of 4 traceroute hops, including start and end) takes 3 to 4ms. At 0.6c, even assuming we're both wired to the core of downtown, only 0.4ms of our RTT is actually spent on speed of light. The other ~3ms is spent on packet switching and protocol conversions. Admittedly, all of these latencies are sufficient for most things that I want to do on the internet, and most users don't start seeing issues for most tasks until they're up in the 100ms+ range. That said, I think there's still more to gain here than from straight up throughput improvements.

oneplane 2y ago
The latency is also where the bottleneck of really novel mass-market network applications sits, most of the time. Just looking at examples like: https://gist.github.com/jboner/2841832 https://gist.github.com/jboner/2841832 for a network to be usable as a somewhat fast SSD, you'd need a maximum latency of 0.15ms. And at that point all we have gained is a network that can be somewhat used like a 10 year old SSD. To then get to a really relevant latency we have to get to memory reference speeds, 100ns at most. That's 0.0001ms or something like that (might be off by one order of magnitude). But with a 0.6c limit, we're not going to get there. Like the other comment, having everyone get a much better minimum speed (say 500Mbps) and low base latency (say 1ms at most to the first IX you can get to) would make a huge difference. But to reference yet another comment, you'd probably just get 4K video ads and larger frameworks with less optimisation since the bandwidth takes care of it anyway.
h_tbob 2y ago
Haha this made me laugh. If our ancestors could see us commenting… “it took 4 milliseconds… should be faster”. So funny, I don’t understand how you even notice the difference win 4ms vs .4!
- aftbit 2y ago
  Even myself from the 2000s would be amazed by the bandwidth that we can muster today. I am just old enough to remember waiting all day to download an IE update over dial up, and I'm definitely old enough to remember a 700MB torrent taking a whole day to download. Now I download and install a 100GB game from Steam in half an hour. But in that same time, where bandwidth available to me has gone from <100 kbps to >1 gbps, ping has only gone from ~100ms to ~10ms. The legacy telephone system was actually pretty good at latency, as long as you didn't end up on a crappy GEO satellite link, which dial up never would. After all, voice is actually a pretty latency sensitive application. That said, our older ancestors would indeed be amazed at sub-second communications. Heck, even airmail latencies would seem insane to someone from before the first world war.

by aftbit 2y ago

oneplane 2y ago

The latency is also where the bottleneck of really novel mass-market network applications sits, most of the time. Just looking at examples like: https://gist.github.com/jboner/2841832 https://gist.github.com/jboner/2841832 for a network to be usable as a somewhat fast SSD, you'd need a maximum latency of 0.15ms. And at that point all we have gained is a network that can be somewhat used like a 10 year old SSD. To then get to a really relevant latency we have to get to memory reference speeds, 100ns at most. That's 0.0001ms or something like that (might be off by one order of magnitude). But with a 0.6c limit, we're not going to get there. Like the other comment, having everyone get a much better minimum speed (say 500Mbps) and low base latency (say 1ms at most to the first IX you can get to) would make a huge difference. But to reference yet another comment, you'd probably just get 4K video ads and larger frameworks with less optimisation since the bandwidth takes care of it anyway.

h_tbob 2y ago

Haha this made me laugh. If our ancestors could see us commenting… “it took 4 milliseconds… should be faster”. So funny, I don’t understand how you even notice the difference win 4ms vs .4!

aftbit 2y ago
Even myself from the 2000s would be amazed by the bandwidth that we can muster today. I am just old enough to remember waiting all day to download an IE update over dial up, and I'm definitely old enough to remember a 700MB torrent taking a whole day to download. Now I download and install a 100GB game from Steam in half an hour. But in that same time, where bandwidth available to me has gone from <100 kbps to >1 gbps, ping has only gone from ~100ms to ~10ms. The legacy telephone system was actually pretty good at latency, as long as you didn't end up on a crappy GEO satellite link, which dial up never would. After all, voice is actually a pretty latency sensitive application. That said, our older ancestors would indeed be amazed at sub-second communications. Heck, even airmail latencies would seem insane to someone from before the first world war.