[shoguning]

The latency is only reduced from tower to phone right?

The tower still has to ping the rest of the network to find the service you were requesting to in the first place.

If total latency ~ 500 ms, phone-to-tower is maybe 50ms of that in 4G. Even if it's down to ~1ms, total latency is only down 10%.

Not a networking expert so maybe I'm missing something here. Perhaps better use of CDNs could make this much faster?

[markalexander]

Computing is getting closer and closer to the edge. One example, as you said, is putting it at the CDN PoPs, but even beyond that you can put your compute into the CSP locations.

E.g. https://aws.amazon.com/wavelength/, but if you’re big enough I’m sure you can make more interesting deals.

[ignoramous]

https://vapor.io/ might be building out the edge network faster than Amazon. Cloudflare is a customer.

[stephen_g]

I would expect the last mile to generally be one of the larger contributors to latency unless you're traversing very long distances (i.e. cross country, undersea cables, etc.). For example, just looking now I get 23ms to a close speedtest server on my LTE and 3ms to the same server on my work's fibre connection. Going a larger distance (~1000km to the next large capital city) it's 43ms vs 18ms on the fibre. That's quite a significant difference, but not as bad as it used to be (3G could have hundreds of milliseconds of latency from device to tower).

[gridlockd]

> If total latency ~ 500 ms, phone-to-tower is maybe 50ms of that in 4G. Even if it's down to ~1ms, total latency is only down 10%.

How did you come up with 500ms for comparison? A fast-paced videogame with that sort of latency would be unplayable. 100ms is still considered bad and 50ms is considered acceptable.

[shoguning]

I guess I was thinking of initial webpage load, which might include a bunch of network calls (DNS, CSS, JS resources).

Each individual network call is faster tho, good point.

[simonebrunozzi]

> phone-to-tower is maybe 50ms

Maybe 50 microseconds, not 50 milliseconds. 1/1000th of what you assumed.

Towers are typically up to a few miles away in dense areas; they don't work well beyond 15-20 miles from the cell phone.

Waves go 186,000 miles (or 300,000 km) per second. Assuming a 10 miles distance from phone to station, it takes ~25 microseconds to do the trip, which is ~50 roundtrip.

So, in other words: latency is NOT affected by 5G much.

[beojan]

That's assuming latency is entirely due to propagation delay, in which case obviously a change in technology wouldn't help. Latency isn't entirely due to propagation delay though.

[vardump]

4G phone to tower latency is about 10 ms.

[HALtheWise]

It seems to me like Uber doesn't depend on high bandwidth or sub-second latency.

[ric2b]

It's high bandwidth compared to the standards of the time they're discussing, not by the standards of today.

[rasz]

Uber would work just fine using WAP/GPRS.