1.But what amount of transaction per second would this deliver?
2.can this really compete against LN in terms of being able to be at the same level of VISA and the like?
1. no, its not about tx per second. its about relaying speeds of solved blocks to clients imagine it this way.
a pizza is 1mb and oneside has pepperoni, the other has anchovies.. the proposal is to cut the pizza in half but you as a full and hard worker(full node miner) still eat the same 1mb whole pizza.. thats right the full 1mb limit nothing has changed.
but because its in 2 halves, its now easier for your witness to CLONE(relay) just the pepperoni half of the pizza because he doesnt like anchovies(signatures). and thus he as a separate thing holds onto just 500kb pepperoni of his copy of the pizza
so each 10 minutes you(full node/miner) are still eating a whole 1mb pizza and your witness has cloned a copy of the pepperoni side meaning he is only eating 500kb per blocktime.
for the full nodes mining the real bitcoin blockchain. nothing has changed, the data totals are the same and thus no extra transactions can fit into bitcoins blocks. the only difference is that the miners data has a split in the middle to make it easier to relay a copy of a half to the network of lite clients..
its not changing the transactions per second. and is only reducing the data per block for the witness(liteclients) who dont want to download everything.
2.witness wont scale anything.. its just a tool for SPV wallets to not be as bloated.. nothing to do with the real blockchain that gets mined properly
LN has more potential to scale however i have seen a few limitations that dont make it perfect.
This is all 100% true in itself as Segregating the signature isn't intended to increase TPS by itself, but with a soft fork core can increase the witness data set limit to 4MB thus indirectly increasing TPS to 6 to 28 tps. It isn't the Segregation in itself that increases the TPS but the partial and limited blocksize growth that only applies to part of the data set. Thus after the softfork there will be a 1 and 4MB limit concurrently depending upon which dataset.
Full nodes will take on more data, but lite nodes will have new features like fraud proofing (although this isn't as thorough as using a full node) and take on as much data as the blocksize limit never increased and have 6 to 28 tps.
There will than be 3 categories of nodes-
1) Traditional and the least secure SPV nodes
2) Lite / Prune nodes with SW nodes that have some fraud proofs
3) Full nodes
Ok, that doesn't seem so bad at all. It's just a sidechain solution.
Well it was tested as a sidechain in elements alpha , but this is actually intended to be an immediate soft fork on the main chain.