4 or 8 to start. I submitted a proposal to the conference (which was rejected), 4-8-16-32, doubling every year. If it started at 8, it could double every two years.
I'm a firm believer that in the short term, it would be really stupid for us to cap user growth. It would be stupid for us to cap user growth in any time frame, but there are other scaling solutions that will work or are highly likely to work, but they are one, two or three years away from being implemented. That is the gap we need to bridge.
I am completely comfortable capping at 32MB for now (for the foreseeable future), and to understand why, lets look at some rough numbers.
32-MB would allow for 60-Million tx per week, and 3-Billion tx per year. I see the Lightning Network as a certainty, but one that is understandably taking time to implement. Lightning will have a multiplier effect, so that each on-chain transaction will be able to back some greater-number of lightning transactions. What is unclear is what this multiple will be, and the multiple itself will like grow with time as the lightning network grows and matures. It could be 10-1 or 100-1 within a few years of being implemented. So in a sense, the total bitcoin transaction rate could be as high as 1,000 tx/sec or 10,000 tx/sec using those two multiples as examples. That is based off the idea of a 32-MB block yielding 100 on-chain tx/sec, multiplied by 10 and 100.
Even though the number of tx/sec could essentially be unlimited with lightning, the number of USERS who have access to the blockchain would effectively still be finite. Lets imagine Lightning taking on the role of today's credit cards. Users would need to settle on occasion, at least monthly. 32-MB blocks would allow roughly 200-Million people to each make a single monthly transaction. Considering the fee market under such conditions, the idea of paying $1 for a monthly settlement transaction seems reasonable, though perhaps on the expensive side, and $0.33 per tx shouldn't dissuade any phone / computer user from participating. Those two fee rates would generate $3-billion and $1-billion per year in total fees, respectively, which will go a long way towards paying for network security.
Bitcoin-NG. I've read the paper (and understand 95% of it). I think it will work, and it solves a few of the trickiest scaling problems. The Bitcoin-NG system eliminates miner-bandwidth differentials as a source of centralizing pressure. Single-packet latency sill matters, as it does currently, but such is unaffected by block size. With NG, it becomes reasonable to use a significant fraction of a user or miners total available bandwidth and processing power to process blocks. A 20% capacity on a 20-Mbit/s connection can yield 30-MB / minute, which would give the equivalent of 300-MB blocks under the current system. Such would have a native, on-chain capacity of 1,000 tx/sec, for 600-million monthly users, who should each be able to perform 10's to 100's of Lightning transactions per month.
And Sidechains... they hold great promise as well, but they are not as well defined as Lightning or NG.
Hopefully we can all see that 32-MB blocks can get us very far when combined with Lightning, so it should be safe for Bitcoin as medium or long-term cap Bitcoin-NG will allow Bitcoin blocks to effectively be some multiple larger than they otherwise would, say 5x to 10x. In the short term, we must expand capacity to grow the system. Choking off user growth is the biggest security risk to the network. Big blocks have their own issues, but 4 or 8 MB would not materially change the centralization pressures today, and 32-MB will not either, say 4 or 6 years out, as electricity costs continue to dominate the equation.
you seem to have a handle on this....what happens when there is a paradigm shift in electrical generation....granted not 4-6 years out ..but maybe in 20 yrs or so when nuclear fusion could be a reality.
Here is the link to the paper.