A new spot market makes cloud computing a commodity
economist.com
economist.com
Practically speaking, to be properly fungible, it should be possible to move compute workloads across the 'Net into whatever environment happens to be cheapest. But for anything being done with large amounts of data, there's a non-negligible cost to moving that data across the Internet, and this in turn can lead to things like lock-in. In fact one could argue that it's almost the only form of lock-in practiced by AWS.
On the other hand, if compute power DOES become a commodity, that would be great, because then Jevons paradox would probably kick in, which could itself lead to true ubicomp etc. One can dream :-)
If all electric heaters in the world provided CPU power for us as well, where would we be?
If a workload is truly distributed, simply fetches data blocks to process then goes computing for a while (Example: folding@home), the amount of bandwidth doesn't matter.
Also, some people have substantial amounts of bandwidth to their house :)
Search http://cseweb.ucsd.edu/users/swanson/WACI-VI/ for "home water heaters".
Not much detail on this. The computers are "probably" doing something part of the time? Really? So when the animation crew goes to use their machines they have to disable them from the commodity network first? What happens if scientists are in the middle of using them? Are the hard drives sometimes unexpectedly full of porno and pirated Mp3s? Is the phone company going to notice the 10000x increase in the data coming in and out of these machines? Are any companies storing personal data on these XYZ random machines?
This is a cuter equivilent though:
http://www.pluraprocessing.com/
It steals cpu cycles from people playing online games.
I imagine this software is similar, which is why it's not meant for people that need reliable access. If somebody starts rendering in the middle of a SpotCloud job, then the SpotCloud job doesn't continue until the rendering is done. And if the rendering doesn't finish until after the job expires, then the job is restarted on a different computer. The worst thing you could do (from SpotCloud's perspective) would be to have your computer available for 99% of the time it takes to do the job, and then become unavailable until after the deadline, but I imagine that doesn't happen very much.
This doesn't make much sense to me. Google App Engine is not very "global" and its bandwidth prices are pretty standard unless you're small enough to fit in the free tiers.