His CPU: i7, 2.8 GHz
App Engine frontend CPU: 600MHz
He does a test and finds that his CPU is 5x faster. 2800 MHz / 600 MHz = 4.67, so his findings match the documented instance speeds and yet he is surprised.
His pricing math is all wrong. Instances live 15 minutes after the last request and you are charged for that. He claims that the amount of work he can do on his i7 CPU for one hour would cost $8 on App Engine. Taking into account his researched 5x faster speed, one hour would result in 5 hours on App Engine, 5 hours + 15 minutes after the requests would mean 5.25 hours * $0.08 = $0.42, which is far less than his $8 statement.
One GAE instance is 5x slower than your CPU, period.
It starts up an instance for a minimum of 20 minutes, so you will be charged for 20 minutes whether you generate 1 RSA key or 120.
Instances don't have a minimum runtime, they have an added runtime of 15 minutes.
Or to get even more technical, you are billed for an added 15 minute runtime but the instance can still be shut down. However if an instance is needed within these 15 minutes and the last one was shut down, then a new instance is created and billing-wise it will act as if the old one was running all the time.
An example to illustrate:
T+00m request comes in, instance #1 is created
T+05m request comes in
T+07m instance #1 gets killed due to google spilling coffee on the server
T+10m request comes in, instance #2 is created
T+25m instance #2 has been idle for 15min, gets killed
-
Total charge: 25 minutes
His pricing math is all wrong. Instances live 15 minutes after the last request and you are charged for that. He claims that the amount of work he can do on his i7 CPU for one hour would cost $8 on App Engine. Taking into account his researched 5x faster speed, one hour would result in 5 hours on App Engine, 5 hours + 15 minutes after the requests would mean 5.25 hours * $0.08 = $0.42, which is far less than his $8 statement.