optimize startup performance

[azahnen]

During every startup some computations and generations take place, which can become really slow for big datasets and/or many APIs.
The goal of this issue would be to reduce the amount of times these computations are executed to a minimum by introducing caching mechanisms or similar means and only execute them when the source values have changed.
A special case of this are multi-node setups where every node executes the same computations at startup. So any solution needs to support sharing of the computed values.