with anything new like this, i'm interested in knowing the upper and lower bounds of the IBLT data size.
lower bound: theoretically, as the UTXO set difference shrinks to zero with 0 network latency, the IBLT will shrink similarly but can never reach zero since it has to at minimum relay enough data to convey the exact subset of tx's included in the miner's block. how small can this data get esp in relation to the avg block size now?
upper bound: if the UTXO set difference is 100%, how big does the IBLT data size get? or does the entire IBLT concept fall apart at some intermediate set difference?
OK, so the bounds.
Note that UTXO is unspent tx outputs held in the blockchain, but
unconfirmed tx are in node mempools, pending inclusion into the blockchain by miners. What Satoshi achieved with his PoW blockchain breakthrough was consensus on the whole transaction history, and resulting UTXOs. I think of that as the steel-work in a skyscraper. What IBLT does is go to further by enforcing consensus on unconfirmed transactions, it pours concrete into that steel reinforcing. This is a significant improvement to cryptocurrency, but it is truly effective only at high volumes, volumes too high for the existing standard method of block propagation.
The reason is that right now the whole network could agree on the same 2000 unconfirmed tx, real-world business, but the next block mined can contain none of them. It could be full of previously unknown gambling dice-bot spam tx, a set which is 100% different. Because these tx validly spend UTXO, then the block is accepted, and the 2000 unconfirmed tx have to wait for the next block. However, when blocks are too big to propagate in the existing method, and require IBLT, then miners have to rely on the majority of the nodes knowing their unconfirmed tx in advance, they can only include a few unknown tx as the more unknowns they include the more likely their IBLT will fail to get decoded, and it gets rejected.
I expect IBLT blocks to be a constant size, starting at about 500KB or 1MB and staying at the same size for a long time, then increasing in steps, perhaps doubling after a number of years, just as the block reward halves periodically. A 1MB IBLT contains about 1500 differences, which means a 1% difference in mempools allows for 150,000 tx or 60MB disk blocks. A 32MB IBLT could support 3GB disk blocks, or 20,000 TPS, as by then differences of <1% should be the norm.