A Cluster Of Pocket PCs

Don Stratton writes "This is the coolest thing I have seen anyone do with a Pocket PC... ever! Well-known Pocket PC developers SPB Software House, located in Russia, have come up with a very interesting spin on computing clusters. The short version is they connected 12 Pocket PCs together in the first known 'supercomputer cluster' of its type and had it calculate the old '3n + 1' problem. It was just done for fun, and not intended to seriously compete with desktop computers, but it does point out some interesting possiblilities for the future of handhelds with wireless connectivity working in ad-hoc computing clusters."

Comments

this only says...

[mantera]

expect things to come out of russia, china, india, and eastern europe in years to come.

Wireless Clustering

[Rich0]

Come to think about it, having something in between WI-FI and bluetooth might make for big wireless networking capabilities in a handheld. If the market penetration were high enough, you could route a packet from anywhere to anywhere as long as you had a high enough TTL.

The challenge would be in organizing the routing tree. You'd have an advantage in that generally two nodes that are close to each other would tend to stay that way over moderate periods of time. Even on the road your handheld would stay close to the others in your same lane of traffic.

If you could set up these devices to be able to share CPU automatically when idle it would mean that your handheld could utilize the CPU and RAM of the handheld in the briefcase belonging to the guy in front of you on the plane.

Certainly none of this is ready for prime time, but it does raise some interesting possabilities.

Wireless or not...

[SharpFang]

...a "cluster" of more or less randomly distributed and connected computers isn't such a bad idea.

On universities, 99% of computers run with nearly zero CPU load for most of the time. People read emails, surf the web, but for most of the time the computers idle. And then someone has some work reaoznajeszdy, waits in queue for a month, throws data on the university campus dedicated cluster, waits for results for a week and receives results that are invalid due to some mistake in input data, so whole procedure must be repeated all over.

Now imagine, we install a "cluster server" on all networked computers. Assign certain resources to the project and let our PC participate in that cluster. It loads a custom computational module for given task, loads data from some anonymous dude on the other end of the world and computes his project. Heaviest "daily" stuff gets finished within few hours. It doesn't really disturb you - works as "idle task", just like SETI@home or such. But, say, you're a raytracer. You prepare a nice animation in LightWave and would leave it overnight to render. Just upload it to the net and have it rendered in 5 minutes on the worldwide cluster. Cool, eh?

Of course the system could be abused. I think some "credit system" would be in order, so people who provide more, get better priority. Plus some way of authoring the "modules" so it couldn't be used to take over the computer. And of course this would be the first step to creating a self-conscious AI, good or evil :) But I think it would be worth a try.