I've been a heavy App Engine user since the beta days -- even gave a talk at I|O 2009 about scaling with App Engine. I have clients that run complex GAE sites that handle millions of daily requests.
I somewhat disagree with the author about scalability. There is a very narrow sweet spot of apps for which App Engine is a quite natural solution. If you're in the sweet spot, you'll probably scale well without too much up-front engineering investment.
The sweet spot _is_ narrow, though. For example, as the OP states: geo data doesn't really belong on GAE -- you can do a limited set of bounding box/proximity queries with the third party geomodel library, but wow, it's expensive and dog slow!
I do agree with the author about both the status dashboard (it only sometimes reflects my current experience with the system) and the surprising variation it data store latency. Latency has been much improved of late with 1.4 and beyond.
at google io 2010 they discussed "next gen" queries in depth, which should be able to do geospatial queries easily (or any query for which your search space can be filled by a space filling curve):
I must be in it, because I have rarely seen a downside to using AE. Here's the nature of my apps:
1) Extremely simple or very flattened data-model
2) Few writes, TONs of reads (I peak at 150 requests/second every day)
3) The occasional DeadlineExceededError causes me little or no headache. For some people this would be frustrating.
Also AE is awesome as a CDN. The latency is very tolerable for static assets.
I somewhat disagree with the author about scalability. There is a very narrow sweet spot of apps for which App Engine is a quite natural solution. If you're in the sweet spot, you'll probably scale well without too much up-front engineering investment.
The sweet spot _is_ narrow, though. For example, as the OP states: geo data doesn't really belong on GAE -- you can do a limited set of bounding box/proximity queries with the third party geomodel library, but wow, it's expensive and dog slow!
I do agree with the author about both the status dashboard (it only sometimes reflects my current experience with the system) and the surprising variation it data store latency. Latency has been much improved of late with 1.4 and beyond.