So, a company like Google provides the software framework, and a “cloud computing” service to clients who have the need for quick access to computational power, but no infrastructure of their own. A game producer makes their game compatible with the framework, making it possible for Google to send jobs to player’s machines. The broker also provides necessary privacy to clients. Probably the biggest fear of James B. the aspiring businessman is that Jane D., his arch-nemesis, will somehow intercept his precious accounting data.

The result is that Google makes money and the game’s producer gets a share. The players pay for the game with a spare resource they don’t need at the moment, rather than actual money. And they don’t have to watch any commercials. The clients get their jobs done.

The upside is the opportunity for a game’s author to make some money, while the presence of a broker makes it easier for everyone to trust that their computers won’t be used for calculating ballistic trajectories, for example.

The downside is that, usually, when a powerful broker is involved, content providers get screwed in the long run. Recently, it’s happened in the Flash games segment – middlemen get some 80% of the advertisement income while content providers barely make a living.

Another danger to this idea is that computational power is extremely cheap, so why would anyone buy it anyway.

(then again, if I could data-mine my books for various stuff with the same ease I can make a Google query, then I probably would)

There is also a limited number of computational problems which can be divided into and solved in large chunks, without much communication overhead. But that should become less of an issue, once those 100Mbps lines we’ve been promised become a common reality.

Now although zombie network is a tempting proposition, it is rather immoral and can get your users into trouble. That’s why I proposed a purely computation oriented cloud that does not take over the identities of your CPU donors. Technically this could be ensured in several ways: by running native code in a sandbox, by running scripts with all I/O functionality disabled, running cuda/OpenCL kernels (which do not have any I/O), or by using a fixed, predefined computation and allowing the host to only supply data batches.

It could make sense if you are running a research project like computing large Mersenne primes, but then you don’t need to make a game for it, just ask for the CPU as it’s already done (games could still be used as a marketing tool though).