[tlrobinson]

It's been done (but for SETI, not password cracking):

http://ajaxian.com/archives/massively-parallel-crowd-sourced...

However, this requires the client have Gears installed, which is basically no one. There's no reason you couldn't do it in the main thread, as long as the task were easily split up into multiple chunks and you were careful not to use too much CPU time.

The hard part is getting it widely distributed. The big traffic sites like YouTube would never do something like this. If you're unscrupulous you could inject the JavaScript in any site you can hack (WordPress blogs, etc...)

A MapReduce type implementation would be very very cool. Make it generic, such that you could submit bits of JavaScript code and the URLs of the data to operate on, and the location to post the results too.

It would surprise me if botnet owners hadn't already done something like this. Of course they can run whatever native code they want.

[jamess]

Hah, blast. Just when you think you have an interesting new idea... Certainly not using google gears would be a good idea, though.

[tlrobinson]

It would still be an interesting project. Especially if it were a generic MapReduce type thing.

It's crazy to think how much power a major site like Google could harness if they actually did this. If I were them I'd do it, but make it opt-in, of course.

If it weren't for the same origin policy they could make a distributed web crawler that loads web pages, parses them, and returns a nice clean list of words and links for the real heavy duty processing. They would need some assurances that clients didn't tamper with the data. Perhaps several clients could process the same pages, and they could ensure they match before accepting the data.