@snek any update on your PR? Would it be easier to optimize multisend for local processes on the same node first?
This optimization should be very useful for Erlang messaging brokers and chat servers:
2 Likes
@pablopla, I worked on it some time ago, and more or less had a working prototype for it, but I also saw many edge-cases in the C layer of the BEAM that I don’t have the confidence nor the time to go that deep into — and also not the proper sponsorship, I’m always happy to contribute any low-hanging fruits for free but ensuring everything in this low C layer is absolutely correct is way more demanding that I can afford 
Furthermore, optimise for local processes would actually be a lot more complicated, as receivers have different flags for off_heap/on_heap messages, available memory to insert new message records, and whatnot. I didn’t have the time to go that deep into those dungeons.
1 Like