Block propagation and validation duration during slow-to-validate blocks on Signet

[b10c]

@ajtowns

I wonder what the TCP window size is in practice? I guess there’s no tools like netstat that let you observe that?

There is! ss -tipn '( sport = :8333 or dport = :8333 )' shows a bunch of information on the connection along with snd_wnd and rcv_wnd showing the window size.

@ajtowns

I was more thinking that having a target prefill size of about the TCP window size, then if there’s more block data that would be good to prefill, having each high bw connection choose a different subset of tx data for their prefill would be good, rather than limiting the size per se.

That’s what @davidgumberg has been working on. Since TCP window sizes vary a lot across our connections and might change over time, (my understanding is) the currently favored approach is to cache a coinbase-only prefilled block and a prefilled block (from what we needed to request or use from the prefill we received; up to a certain size) and then decide while sending if the window size allows for sending the prefilled or the coinbase only block. My understanding is that he saw significantly better reconstructions with this.