can anybody tell about TPS rate and predictable database growth?
There is no TPS set in stone as with bitcoin, where the blocksize basically determines how many transactions you can fit in a block, with Byteball, theoretically it should handle hundreds of thousands tx/s, but today its not optimized at all for this speed.
Correct.
I guess the upper practical limit is the 12 witnesses, how many transactions they can see (bot dont need to see it at the same time) and post to each other and append to the database, to the main chain.
Wrong. Witnesses don't do any more work than any other full node, they are not special with respect to TPS, and they are not a limiting factor in any way.
There is nothing stopping you from making really a massive amounts of transactions on your own chain - and have it merged to the Main Chain when posting it for the witnesses to see. As long as you didnt double-spend, the verification of which witnesses do takes little time compared to what you spend in making the transactions, if all is fine, your chain of massive amount of txs will be in the main chain.
Again, this implies that it is witnesses who are doing verification and detect double-spends. Wrong - every full node does it in exactly the same way.
We use witnesses only to establish order of events by looking
back in history.
Wow, thank you for the corrections! Much better than I even realized.
Does adding more byteball-hubs help the network process more transactions? Does a byteball-relay also help, but just doesnt store temporarily any private-messages like a hub does?