Concurrent JavaScript — Part IV: Madness!
In a previous article, I talked about I/O performance. Remember when Nexus.js used to convert 400MB of text from UTF-8 to UTF-16LE in 1.8 seconds and I bragged about it?
That was quite an achievement for me, but then Christian Landgren pointed out that Node.js easily beat that using cluster.fork(), and it finished in 1.178s.
I was quite distraught, until I split the FileSourceDevice implementation into two types of devices:
First, you have the FilePullDevice, which is the old — but renamed — type.
And then I added the all new FilePushDevice.
FilePushDevice relies on events, it inherits EventEmitter and dispatches the ‘data’ event whenever data is available, so let’s try it out:
Madness!
It just converted 400MB of data — outputting 800MB! — in 0.484s! That’s almost a quarter of the time it took before!
This performance is the result of piping the streams in parallel, switching from the iconv library to ICU, and good data flow. I also took the time to tune the buffer sizes, which will be programatically adjustable in a later commit. (Don’t forget to remind me, brain!)
I also have another update, I just started implementing networking, starting with UDPSocketDevice. It can’t listen yet, but it works well as a client.
I’m still pondering the implementation dynamics for binding ports. I’m thinking of implementing an Acceptor class, which will be responsible for accepting connections from a socket.
We’ll see.
If you have any ideas on the matter, I’ll gladly welcome them! I would love to have some input on this subject!
This concludes the fourth instalment in this series. I certainly hope you like how Nexus.js is progressing so far, and as always: you can review the code for the project on GitHub.
You can also follow me on Twitter for real-time progress updates!
