Cispy

Communicating Sequential Processes for JavaScript

Fork me on GitHub

10,001 Processes

This was a nifty little thing that I found on swannodette's blog which demonstrates the creation and execution of 10,000 processes in the browser. I thought it was very cool, so I tried implementing it (with a couple of modifications) with Cispy.

This is 10,001 processes running simultaneously, all communicating over a single channel. 10,000 of these are each responsible for a single element of a 10,000-element array, changing its value every 1-10 seconds (randomly chosen each pass) to a different number from 0-9.

The 10,001st process takes these values from the channel as they come in and collects them in a queue. It does this through an alts call which reads two channels: the one where all of the values are, and a timeout channel that closes every 40 milliseconds. When that timeout channel closes, a new color is chosen (one in a cycle of seven), the elements in the queue (i.e., the ones that have been updated in the last 40ms) are displayed in their appropriate places on the screen, the timeout channel is restarted, and the queue begins to fill again.

Those who have done multi-threading in other languages will know that 10,000 threads would be impossible within memory and CPU constraints. Processes are much lighter in weight though, to the degree that a browser engine can handle them fine.

The changes from the original are threefold:

  1. Channels, by default, can have up to 1024 pending operations queued up before they complain (I don't know what the limit is in Clojure). Operations after that simply don't get queued, meaning that puts don't get taken, meaning that the process whose put is rejected blocks forever (and its digit never changes again). Intermittent browser slowdowns when your computer is otherwise busy will cause this limit to be exceeded, so I've bumped it up to 4096 pending operations.
  2. I added a sixth (non-black) color, because magenta was under-represented.
  3. The original had each of the 10,000 processes choose a random cell to update. I think the idea of having each cell serviced by its own individual process is elegant, so that's what I did.

Here is the code for your viewing pleasure.