Hacker News new | ask | show | jobs
by tgtweak 1961 days ago
No mention of latency improvements?

Seems crazy since oversea transit (tcp & single-channel) is usually latency (or loss) bound.

I would expect it's better than going over public transit and legacy subsea fiber, but it would have been useful to see some comparison tests between POPs.

2 comments

Google invests a lot in TCP congestion control, mainly through BBR. I believe they do bulk transfers with centrally-scheduled fixed-rate UDP transmission. I also assume they have better control of buffers, loss, and queueing algo’s to prevent/control loss/drops.
I'm not sure how easy it is to increase the speed of light in glass without some sort of new breakthrough.
Not travelling through 20 routers in the process tends to help. Again it would be good to get a tangible idea of how much better this is vs just stating the obvious about peak theoretical throughput.