@js the protocol **by design** allows for arbitrary routing algorithms to be used to relay message between servers. by default we use full mesh for simplicity, but we're actively working on better-than-full mesh (c.f. https://youtu.be/DZBvy4abB1o?t=1223 and https://www.youtube.com/watch?v=vimXuCTxV6k etc). not sure who you were talking to. sorry that it feels to have been a waste of time.
@matrix I’m very glad to hear about this! And I’ll happily give it another try once something other than full mesh is available. Also, if there’s something you want me to try to see if it makes things better for me, I’m happy to try it. I already tried workers and a reverse proxy for TLS.
@matrix UltraSPARC T1, PostgreSQL, cache_factor unchanged, Python 3.6.6, OpenBSD‘s malloc.
With the workers, I always saw 100% in the federation_sender, federation_reader and homeserver. After leaving it, things work for a while, then the homeserver suddenly is at 100% CPU while apparently not doing anything and being unresponsive. Restarting everything fixes it, until it happens again. media_repository dies with SIGBUS after a while (wonder if that could be the reason for homeserver @ 100%?)