Thread by @AdamSpitz

1) Various blockchains (#Ethereum, #Avalanche, #Cosmos, #Polkadot, etc.) use a "central thingy with a bunch of sub-thingies around it" model, for the sake of scalability.

They're not all equally secure, though.

Let me try to differentiate the good approaches from the bad ones.

2) (I should also mention that another intended purpose of this model, besides scalability, is allowing more flexibility/experimentation/sovereignty in the sub-thingies, which is cool too.)

3) Ethereum has two such scalability strategies on its roadmap: layer2s and sharding. Both of these have the advantage that the central thingy has a single set of validators that secures all of the sub-thingies, and the system's security *scales* with the number of validators.

4) The insight of layer2s is: it turns out that it's possible (using Cleverness!) to separate "doing the work" from "verifying that the work was done honestly". So we have the full validator set of layer1 Ethereum verifying that the layer2 has done its work honestly.

5) e.g. A ZK-rollup is a layer2 that does its work (i.e. produces its next block) in such a way that (using clever cryptography) it produces a validity proof that it did so honestly; layer1 can cheaply verify this proof. So the L2's validity is secured by L1's huge validator set.

6) The insight of sharding is: it'd be fine to break up the chain into a bunch of separate shards and have each shard secured by only a fraction of the validators, *as long as* we randomly shuffle those validators (i.e. reassign validators to different shards) pretty frequently.

7) That way, an attacker can't just pour all his resources into attacking a single shard; he'd still need to control a huge percentage of the shared validator set in order to have a chance of controlling a majority of one shard (and even then, only until the next reassignment).

8) I'm not super-familiar with @Polkadot, but I believe they're focusing on doing sharding first, whereas @ethereum is focused on supporting layer2s first and is planning on doing sharding (particularly data-shards) next. Both those approaches sound fine to me.

9) In contrast, @avalancheavax and @cosmos have superficially-similar approaches (Avalanche has a "primary network" with "subnets" around it, Cosmos has "Cosmos Hub" with "zones" around it), but use neither of the above approaches for making their security scale.

10) e.g. The Avalanche docs say that all the validators have to help secure the central thingy (which is good), *but* each sub-thingy is only secured by whichever validators want to secure it (which is bad).

11) This (from the Avalanche docs) is pretty damning, IMO: "In a heterogeneous network of blockchains, some validators will not want to validate certain blockchains because they simply have no interest in those blockchains."

That's not a good thing!

docs.avax.network/subnets

12) That is, if your company creates its own Avalanche-based blockchain for the sake of scalability, your blockchain does *not* inherit the full security of the central Avalanche network, the way an Ethereum-based layer2 would.

And IIUC the story with Cosmos is similar.

13) Security is hard to evaluate in advance, because all the technical arguments sound like indistinguishable mumbo-jumbo to most observers, and so it's really easy to ignore it or just kinda assume that it's been taken care of, right up until there's an Incident.

14) I hope this helps clarify the differences between the various "central thingy with sub-thingies around it" scalability approaches.

I think Ethereum is doing it right; security is super-important, and some of these other chains' approaches sacrifice it unnecessarily.

(end)