Depending on your use case, zmq inproc I've found has a lot more latency than you would expect. like p50 of >100us and p99 can be over 1ms. Kind of unfortunate

Depending on your use case, zmq inproc I've found has a lot more latency than you would expect. like p50 of >100us and p99 can be over 1ms. Kind of unfortunate as other more handrolled solutions (depending on use case) can easily be <10us. Feels like you really need to be using a lock free ring buffer that only supports 1<>1 communication if you really want to go fast. (p50 of <1us with very compressed tail latencies)