I'm not sure if you're refuting what I said, but there is absolutely no way you will get 20ms end-to-end latency from this. Satellite propagation delay, yes, but for your ping to get to Google and back will be over 50ms.
They have no cross-links, so up and down to space is going to happen multiple times in many cases. Second, they will not have data centers or ixps right where Google's ingress is, so it has to traverse fiber for quite a while. Someone gave some real numbers on the last SpaceX thread, but I believe that part alone will be 20-30ms added.
They do not have cross-links. They were in the original public announcements, but none of the current satellites have them, nor have they announced when they would be launching some that do. It's likely years away.
That's a claim musk made, but they already have over 500 launched without them. They should have 1000 by year-end, so it's unlikely they'll be able to use them for a couple of years.
Would satellite on-board caching be possible? Could you cache, for instance, the most frequently requested cat photos currently on reddit? You would be able to bypass half of the round trip
Possible, yes, practical, no. In general, you want the least complexity as possible in the payload, since you can't fix it as easily as you can with ground bugs. Also, memory that's space-hardened is going to be significantly more expensive and failure-prone, so you need a lot of redundancy.
AFAIK Verizon is doing something similar with edge compute in their 5G towers/data centers. If Starlink gets big enough, I'd expect them to do the same. However, I think it's much more likely for Starlink to put the edge compute/cache at the Starlink ground stations, rather than in the satellites themselves. I think the tech/cost still isn't good enough to have significant compute on a satellite.
