On zero-copy: the article claims nanomsg uses it. The official documentation at http://nanomsg.org/development-bindings.html http://nanomsg.org/development-bind

On zero-copy: the article claims nanomsg uses it. The official documentation at http://nanomsg.org/development-bindings.html http://nanomsg.org/development-bindings.html says something else: > Zero-copy means that the application has to deal > directly with different native memory buffers, for > example with blocks of shared memory, or memory pinned > down to physical memory address (so as to be accessible > by zero-copy CPU-bypass techniques) and so on. This is > hard, and sometimes even impossible, to get working > with languages with automatic garbage collection. > > If you still want to provide zero-copy, make sure to > discuss it on the nanomsg mailing list or IRC channel > in advance to avoid possible pitfalls. So, I assume that zero-copy is not used in any out-of-the-box implementation. Maybe someone with more experience can clarify this. The documentations further mentions zero-copy only in the local form of scatter/gather (used with vmsplice etc.), which is distinct from zero-copy to the network layer.