Yeah, this seems to be another case of json beeing slow. I profiled sbot-11 and the time is mostly spent json stringifying and json parsing, this is also why legacy and EBT are so close. One interesting thing is how stable EBT is, so I'm quite sure that if we land something like the binary format, EBT is going to be a lot faster.

I think the handshake should only happen once. After that 100k messages gets replicated. Also this should test two seperate sbots connected over a network, but without the variation of the network. You are correct that it would be really interesting to do another test where you are communicating or extracting information from sbot directly over a unix socket. I'm not exactly sure what ssb-client does.

Looking at the data again, one thing that strikes me is how similar the legacy and EBT is on both our machines. Legacy is actually a little faster overall, which shouldn't really be the case. Hmm, another rabbit hole to go explore