[Cryptography] Our leader opines on cryptocurrencies

jamesd at echeque.com jamesd at echeque.com
Sat Jul 20 23:32:05 EDT 2019


On 2019-07-20 12:56 pm, John Levine wrote:
 > Master Card can process 38,000 transactions per second, and Visa about
 > the same.  Good luck doing that with a blockchain, even a closed
 > non-PoW one.

Working on it:

Assume delegated proof of stake, in which the merkle patricia root of 
the most recent block is signed by a quite small number of big peers on 
the blockchain, each representing a very large number of very big 
clients each owning a great deal of stake.  (Ordinary clients are 
insignificant. Ordinary peers are insignificant, but the interests of 
big peers and big clients are aligned with those of small peers and 
small clients, if there are enough of them.)

Then each big peer has to process about one hundred thousand 
transactions per second, assuming we want to replace government fiat 
currencies with a blockchain currency.

So we have a small number of peers, each peer processing the entire 
blockchain, and each peer composed of a large number of shards, each 
shard processing a shard of the blockchain.

Assume the processing is shardable with mutually trusting shards that do 
not suffer byzantine failure.

The computer sitting under my desk, with a consumer grade fiber to the 
curb connection to the internet, could probably process six thousand 
transactions per second running four shards in a single physical 
computer, so each peer is going to need forty such shards running on ten 
such devices, say a eighty such shards running on twenty such devices 
for redundancy and storage, each device having twelve six terabyte hard 
disks.

The shards coordinate by non byzantine paxos on a single hash once every 
few minutes, and the peers coordinate on a single hash by byzantine 
paxos.  The number of peers, and the number of shards, is small enough 
that we don't hit paxos scaling problems.

---
This email has been checked for viruses by Avast antivirus software.
https://www.avast.com/antivirus



More information about the cryptography mailing list