GSO oribt is 35k high, back trip is 70k km , so 1/4 second or around 250ms.
That's an approximately correct latency value
IF you happen to be directly below the sat at the equator. For all other locations (i.e 99% of the population) it will be somewhat greater. For those of us in NA it can be in the neighborhood of 300ms...
However, that's not the entire story... having worked in the satellite vertical-market ISP industry, there are a number of factors when considering IP based communications over satellites:
- A typical IP connection is a send/recv channel (i,e, HTTP request & response), hence you are doubling that latency for typical applications. Observed ping latencies of ~700-800ms are typical
- A new IP connection starts with a 3-way handshake to establish the channel. This implies a nearly 1-second delay before any actual data is passed, for
each IP connection.
- Retires due to lost packets are painful as a result of the above mentioned latencies.
- Typical applications open multiple IP connections. For example, a browser can open a connection for each individual object on a web page, with many being sourced form multiple servers (content server, ad server, click track servers, etc...). Thus there are multiple delays that may all not be satisfied simultaneously.
- Most satellite providers do back-end optimization by aggregating IP connection requests in to single sat frames, and (attempting to) transparently proxy multiple object requests to avoid multiple round trip traversals of the bird for things like a page loads. The latter only works for well-known protocols (like HTTP), and even then not always.
- As the TCP aggregation is effectively a (polite) man-in-the-middle attack, it has the side effect of breaking things like HTTPS and VPN tunnels
As earlier mentioned, sat communications can be very useful for largely non-interactive data payloads like multicast. Having done IP-based video and software delivery via sat, I could easily see this as being useful for firmware or map-data delivery to the cars.
However for interactive uses (Nav, web browsing, web radio channel surfing), I suspect there would be some impact in using geostationary sats. Perhaps the LEO satellites Elon is considering would lessen this sufficiently to make it useful.
One useful point is that the satellite link is effectively a private network until it's routed out to the public backbone. Thus even though today the cars use a VPN over the public network back to the Tesla HQ mothership, a sat link could obviate this need and the the above-mentioned penalty by routing the car data from the ground-station to The Mothership via private terrestrial link or VPN.