This repository contains examples of using Jay McCarthy's zeromq Racket FFI bindings. They replicate Clojure examples from this blog post.

The native zeromq library must be installed for the examples to work. They use the Racket binding to zeromq's zmq_proxy call. As such, zeromq version 3.2.x must be installed.

2. request-router.rkt and response-router.rkt. These two scripts operate in tandem. The first script sends 100000 requests to the a set of responders which reply to the original requests

All examples use Racket's places API. The Racket VM runs on a single native thread. Racket's thread API implements a thread as green threads (implemented most likely as Scheme co-routines).

This causes problems, though, when attempting to use zeromq's reply and request sockets in the same process; these sockets block the currently running thread until the reply and request messages are received on either socket. Using these sockets with Racket's green thread simply blocks the whole Racket process.

As a workaround, the examples use Racket 'places' API. Each place actually runs on a native thread, so spawning off the sockets in a Racket place means that Racket VM that spawns them is not blocked.

Using Racket 'places' also provides an additional bonus. The places are distributed across multiple CPUs.

The disadvantage of a Racket place is that each place is actually a separate Racket environment managed by the Racket environment that spawned the place. Each place therefore consumes as much memory as the process that spawns them. For example, on my machine, a single Racket VM costs around 100MB of memory. But while the request-router.rkt runs, the Racket process balloons to 700MB's of memory. It shrinks back down after the number of requests take place.

The code in request-router.rkt and response-router.rkt operate in a way which is very similar to Racket's distributed-places. Distributes-place are places that can operate on remote endpoints across a network. The zeromq examples, with a few tweaks, could operate similarly, since the sockets operate across an tcp channel.

There is, though, an important different. The messages between zeromq sockets are composed of bytes (Racket byte-strings). In contrast, the data between Racket's distributed places are high-level abstractions such as vectors.

All examples are MIT licensed