,
10 tweets,
2 min read
Read on Twitter
There's an efficient frontier for the decentralization / efficiency trade-off. What does a network like EOS looks like when projected on that frontier? First, as distributed computing people will tell you, it's smart to have 6k+1 nodes. So let's pick 19 or 25 but not 21. (1/10)
Let's got with 25. Second, run a classical BFT algorithm between your nodes. You only have 25 participants, no need to be fancy, go with PBFT. As an added bonus you now have finality. (2/10)
Forget blocks. Proposers just commit the full content of their mempool as a batch as soon as the previous batch is committed. Not going into details regarding the backend storage but it probably looks like a SQL database and not something exotic. (3/10)
Let's now look at the cost. We'll splurge for an r4.16xlarge instance on AWS. That gives us a Xeon Ivy bridge with 64 cpu-cores, 488 GB, 20 Gbps bandwith, 100 TB SSD. We're looking in the vicinity of $50K a year. (4/10)
Note that this is a *substantially* beefier setup than what EOS BPs advertise. We need 25 of those, but let's double it and throw in another 25 backups. We're now looking at $2.5M a year. Let's throw in another $250K * 25 for sysadmins. (5/10)
Let's round it all up to $10M a year for network maintenance. With a 5% yearly discount, the sum of the discounted costs of operating your network is about $200M. (6/10)
This is a useful modeling exercise if we look at tokens as rights to access those resources. Does it mean the total value of all the tokens is limited to $200M in our example? (7/10)
It all depends on whether there are network effects on your first network that cannot be replicated by spinning up another one. What are candidates those network effects? The first one is composability / interoperability. (8/10)
If somehow being on the same platform is super important for applications, value is lost by running them in separate networks. There are however few examples of this and, besides, there are interoperability solutions which can bridge those applications across network. (9/10)
I suspect this effect is weak at best. The second candidate for network effect is commending a monetary premium. It seems more viable if it can be achieved. If there is a market for apps running on mildly decentralized networks, it will be about driving down cost. (10/10)
