Note that the erlang wire protocol and binary parsing are also available for Haskell. So I really don’t think interoperability is an issue.

Of course, for distributed systems, Erlang’s libraries are far more developed than Haskells, though nothing’s stopping you putting nodes of Haskell runtimes across a cluster speaking, say, the Erlang wire protocol.

The channel-filling-up leading to STM contention issue is intersting. Maybe you should be using Chan rather than STM Chans there.

I also know I’ve had problems with early experiments with GHC’s parallel Channels. Channels tend to fill up, as the producer gets scheduled more often than the consumer—or the consumer reliably hits STM retries, so the producer fills memory. Erlang’s VM has a nice default scheduler: it balances reductions in expressions between threads, and it rewards consumption of messages from channels. I don’t know how to imitate that behavior in Haskell—how to write one library once and have the scheduling behavior I’d like.