ZeroMQ: Connect ALL The Things

Writing applications in a multi-core, multi-threaded, multiprocess, networked world means communicating between many threads and processes over shared memory, IPC, and the network. This often involves multiple low-level libraries (e.g. most languages’ built-in threading, unix IPC, Berkeley sockets) with different programming paradigms, and may require a potential a scaling bottleneck in the form of a central server or broker to make it all manageable. ZeroMQ claims to be a better alternative, providing a single, higher-level message-passing toolkit across threads, processes, and networks, and languages, and specifically supports decentralized messaging. That should make it a slower, clunkier compromise for any one task, but it claims to be both better and easier to use for any one of those problems than a dedicated library. Extraordinary claims require extraordinary evidence, so we’ll examine how it compares to standard threading and networking facilities and see how easily we can just connect all the things like Lego bricks, regardless of type or underlying transport. Of course it wouldn’t be any fun without including some very informal performance smackdowns.