"hope it will be able to process as many as 10,000 transactions per second, which far surpasses Visa’s capability to process around 1,700"
I think somebody got bad info about VISA's capabilities. 1700/tps is probably their average tps over a day...solely based on needs/demands...not what it could do. Not the peak, and not the edge of their performance envelope. It's likely they can already handle much more than the proposed 10k/sec of this sharded block chain.
Edit: Yep...pretty sure someone started with "Visa does roughly 150M transactions a day" then did 150,000,000 / 24 / 60 / 60 = ~ 1736
I applaud this effort as an academic project. Hopefully it will help move forward the state of the art. But I reject the idea that a new cryptocurrency will "fix the bitcoin problem" by just being faster.
But there are already several blockchains that promise high transactions per second such as Neo, Eos, Nano, etc. And Ripple, which has been around significantly longer than these, claims to be able to handle 50k transactions per second.
Cryptocurrencies need a lot of adoption and real world use to be successful. Beware of premature optimization.
Prediction, sharded blockchains will never take off. With newer blockchain protocols like MimbleWimble, you do not need to store the full transaction details for all time.
Too many things to go wrong, too brittle...
This is a much better scalability technique on a single blockchain -- taking advantage of zk-proofs and EC point encoding. If you want distributed computation also, ala Eth/EOS, you could combine a zk-proof of computation with the MW tx encoding with no loss of generality.
This article does not explain why sharding the history reduces the time it takes for a transaction to end up in a block. It also does not mention that this delay is exacerbated by the standard practice of waiting for several blocks to be mined on top of a transaction to reduce the chances of a fork that renders it invalid.
For anyone interested in more information, I found the research paper and website for the currency named "Unit-e". The paper seems to be quite extensive at 168 pages long and covers a lot more than the blockchain design.