I am not sure whether the comment in PR 27086 I linked is referring to congestion issues or IPv4 fragmentation issues. I don’t have hard data, but I believe both contribute to latency issues here and sending data >> MTU (~1500 bytes) is going to lead to lots of fragmentation. Two links if you have the time:
I’m not really sure that pre-filling above MTU is worth it after reading the two above RFCs, but curious to hear thoughts.
EDIT: Sorry to cross-post, but I’ve TLDR’d the above two RFC’s in a related Lightning conversation here: Latency and Privacy in Lightning - #13 by Crypt-iQ
I think I’ve actually conflated IP reassembly with TCP reassembly. I think maybe hard data would be nice to have here?