Hi, While you are polishing the new protocol, I've got few new thoughts: 1. You are often speaking about implementing routing in hardware. Isn't it easier for hardware to have the whole multi-part message to be length-prefixed, instead of having complex state machine with "more" flag? Making labels fixed size (without length-prefix), as they are currently implemented is also an option. 2. Is the REP socket eligible to respond for SURVEYOR responses? The use case is following: Node A is a surveyor. Node B is responder. It has two processes B1 and B2 which are equally right for responding the survey. The obvious way to implement this currently, is making a device B0 which talks to A by "survey" pattern and speaks with B1 and B2 with "request-reply" pattern. In this case when scaling node B from a single process B1 to a device with multiple processes, you must change the pattern of the socket (read change the application code). Which seems not the best thing to do. BTW, the same occurs with pub-sub and pipeline. There are many use cases where you want pub-sub between nodes (or subclusters) and do push inside. But with pub-sub the issue is limited because of the subscriptions.