As always, Vitalik’s articles are super interesting and tightly-packed with good information:https://vitalik.ca/general/2022/09/17/layer_3.html
One of the uses he briefly describes, for L3, is ” L2 is for general-purpose scaling, L3 is for customized scaling.”. Though it might sound simple at first, I think this phrase hides deep and important implicationsCustomized scaling is likely the most promising use of L3. One can enable exceedingly higher computational loads with application-specific rollup chains because of local consensus. Such application chains (in contrast to traditional sidechains) would use the general-purpose base layer (L2) for dispute resolution, thus deriving the security of the base layer, through a 1-of-N security model.
The shift from global to local (or application-specific) consensus affects how DApps relate to each other. This design might come at the cost of higher friction to composability, but that is a concern that is irrelevant for a broad class of applications.
On the other hand, applications gain the ability to function without competing cannibalistically for the computer power of validators. It is possible to avoid network gentrification and ensure that things outside DeFi survive.
In particular, applications that run on their own optimistic rollup chains with interactive fraud proofs (as compared to ZK rollups) can support computational loads comparable to mainstream (e.g., involving billions of instruction steps and large memory address spaces). It would enable a new class of decentralized applications that use alternative VM architectures running complex decentralized logic with mainstream software stacks.
Being able to customize the app chain and choose the profile of validators is also a plus. Maybe everyone that cares about a cool decentralized MMORPG has a beefy computer that can process a bunch of stuff. You also get to customize the sequencer per application, maybe a DEX could forsake market maker fees for computation.
Thoughts shared by https://twitter.com/erickdemoura on twitter.
What do you think? I totally agree that we’re overusing global consensus and neglecting this idea of leveraging the 1-of-N security model to “localize” consensus and customize scaling.
