I don't understand how the instance-hours are being calculated. One of my apps has the old 2.5 CPU-hours/day to the new 34 instance-hours/day calculation. The app is not CPU-intensive and 2.5 hours of work sound reasonable. It certainly is not hogging the CPU 34 hours the whole day. Does the new instance-hour include the process idle time of app?
They calculate how many minutes an instance is active and sum all that. If your application is network/API/I/O bound, you will get highly inflated counts. However, if you take multithreading into account, one instance of an I/O bound app should be able to serve many, many requests, so your usage will go down tenfold again.
That is, unless I got something very wrong, in which case we're all screwed.
So it counts idle time. When a process is idle, it doesn't use the CPU, which can be used by other processes.
This basically punishes IO/network bound apps and favors CPU-intensive apps.
I can see why some of the web apps has 100X cost increase since they spin up a process to handle a web request which can take fair amount of network time but uses very little CPU.
The new pricing model is not web-app friendly but favors heavy CPU-intensive backend processing kind of apps.
Not really, if you enable multi-threaded request handling. While one thread/request is waiting for IO/network to complete, other threads are still handling requests. All in the same instance.
