it'll be a while yet. we've been focusing on hitting Matrix 1.0 by iterating on the Python codebase (Synapse) rather than diluting the effort by doing it on Dendrite in parallel. Once 1.0 is safely out the door and the first wave of post-1.0 features and perf have landed in Synapse then I suspect Dendrite will see more love.

tl;dr: Synapse is getting (much) better; i wouldn't wait for Dendrite but give Synapse a go on Py3 instead for now.

that said, once Dendrite lands, it should be 1-2 orders of magnitude more efficient by most metrics :)

Thanks. I think you're doing the right thing.

Note that Dendrite is doing a bit more fancy architecture. https://github.com/matrix-org/dendrite/blob/master/DESIGN.md -- there's a bunch of readers and writers, with Kafka in-between to mediate flow of messages.

Whether that's a good or bad thing remains to be seen, but I'm a bit skeptic towards complicating things from the get-go.

FWIW, I'm running Synapse on a 1/1 instance (1 core, 1 GiB RAM) alongside a bunch of other services, with 4 active (as in: daily) users, and it works reasonably well as long as people don't join large channels (cough #matrix:matrix.org cough). That said, I'm definitely looking forward to Dendrite.