And as a number of people are gleefully pointing out, Appengine comes with a strategic cost that is not trivial.
After playing with the .zip file they provide, I think fears of lockin are overblown. It should be relatively trivial to port existing apps to GAE, and apps developed for GAE should be easy to port to a WSGI enabled server.
Google will be a good place to run the free version of your web app that serves as an ad for the premium version that does the same thing with crypto, or with C extensions that add value to the transaction, you'll be able to use the same core code in relatively similar application containers.
I'm definitely going to use it for that. And if I can get away with it, I'm going to try doing some unclever implementations of semi-supervised ML, maybe a few matrix calculations, and see what I can get away with before they kick me off their grid :-)
It occurred to me that, since no one is paying me to optimize the usual tight loops that crop up, or write a C extension to speed them up, I can try just running a shit load of them in parallel and storing the results in their ObjectStore. Now I just need to randomize the ''request'' IP blocks and we're all set for the billion or so simulations that my officemate wants to run.
I understand these articles are written in a rush, but this is plain wrong. From Google's site, it seems pretty clear to me that this will be true out of beta:
Every Google App Engine application can use up to 500MB of persistent storage and enough bandwidth and CPU for 5 million monthly page views.
I'm surprised nobody has mentioned this - does this mean the tech cost of a web-based startup is now $0?