Map-Reduce in JavaScript, working on client Browsers

This kind of ties into a proof of concept I worked on a while ago.

An advantage of the native SETI-at-home clients is that they have access to the gpu. But theoretically so does javascript via WebGL so I took a stab at implementing matrix multiplication entirely on the GPU from within the browser. I think it could be a pretty powerful idea if it were fleshed out a bit more.

http://learningwebgl.com/blog/?p=1828

There was something similar to this posted a few months ago, http://maprejuice.com/. I believe it also used node.js and web workers in order to do distributed map reduce operations.

I wonder what kinds of workloads this is actually useful for?

JavaScript performance has vastly improved in the last few years but it's still relatively slow, as are most individual's network connections.

IIRC a bunch of people tried this approach on the Engine Yard SHA-1 hashing contest, and were obliterated by those using CUDA, etc.

Similar to this? http://www.igvita.com/2009/03/03/collaborative-map-reduce-in...

Don't have time to verify this until the weekend, but if it works as advertised, this is brilliant.

Very cool proof of concept. The advances in web-front technologies are all so very exciting.

If you want to do some actual high-performance work, I'd look at Java applets. Java runs around 5x faster than the same JS code (V8), and much much faster in browsers like IE. Plus they have real multi-threading.