if I asume only serval hundreds of tps then I am wondering how long my data transaction stays alive/visible on the tangle before it is “snapshoted” down. To use permanodes is a way back to centralization. Should the network be seen more like a transport layer which connects 1:1 or 1:n or n:m entities or shoud it be seen more like a database where I can see my data laying around over at least days (e.g. needed for productions chains)?
After coordicide, data can be snapshotted away after about an hour. However, users can choose to keep the data for as long as they want. Moreover, users can delete the data they dont want to save: the data in a message can be reduced to a “merkle hash” which allows to keep a record the structure of the message tangle, but forget the underlying data ).
I agree, the databases will can get large, and we must balancing the ability of the DLT to hold lots of data without restricting access to it. However, ultimately, we will tackle this question when we deal with sharding.
Your next question is quite interesting, because the answer is both. On the one hand, the message layer really is like a communication layer, however proof of the communication will last forever. Indeed, with a merkle proof, you can always proove that your data was entered into the message layer, and, since the timestamps are enforced, you know when (approximately) it was added.