Thread by @epolynya

Execution sharding is still kind of monolithic, or as I call them, proto-modular. It was a necessary step, but rollups & DA layers supersede them. I agree NEAR is the best execution sharded chain yet: though I'll point out "ETH 2.0 sharding" no longer exists as John suggests.

https://twitter.com/j0hnwang/status/1477448481518546944

With the new sharding design proposed by Dankrad, we will likely never see fraud-proven execution shards on Ethereum. So why is Ethereum abandoning execution sharding? Because fully modular architectures like rollups + data sharding is simply superior in the long term.

Rollups+DAL have several advantages over sharding:

1) Retains atomic composability even if it settles across multiple data shards, while execution shards break composability between each other. So, you can have a super duper rollup with gazillion TPS with full composability.

2) Far more flexible - with execution sharding you are restricted to the rulesets. For example, no one's doing ZK-shards yet, AFAIK, and you're restricted to fraud proofs / optimistic shards. We've had ZK rollups for ~22 months now, and even smart contracts live with StarkNet!

To be clear, we will see zk-shards years down the line, but you're at the mercy of the base protocol supporting them. With rollups, you program your own destiny.

The above flexibility is not just about ZK, but also VMs, programming languages, sequencer models, fee models etc.

3) Sovereignty: Further, with rollups, you can build your chain entirely independent of the base protocol, while with shards you're forever part of the protocol. I'll add that I believe secured rollups with settlement are the superior solution, but sovereign rollups are possible.

4) Compression & scalability: with rollups, full nodes exist independently, and by splitting settlement & DA you only need to commit compressed data to the DA layer(s). End result, you get far higher scalability in the 30x-100x range than just execution shards.

5) Dedicated DA layers: I'll note that NEAR uses some of these concepts even for execution. But separating execution state & DA means you can have dedicated DA layers which need not hold any state, and can be easily parallelized. With DAS, you only need to add more nodes...

...to add more scale. Further, with a stateless DA layer, you can put cheap hard drives to use, so you get massive scale at low cost, and the only bottleneck is bandwidth. Also, note rollups have pre-compressed data, so you get more out of your DA layer's increased capacity.

I have great respect for the NEAR team, and those who have followed me since 2020 on other avenues will know I consider it the top alt-L1 project alongside Mina. The key reason being both NEAR and Mina are best placed to transition to the modular paradigm. Most other L1s NGMI.

Short term, while rollups & data shards are maturing (I still stick by my ~2023 expectations), I believe Aurora & NEAR are among the best solutions as the ultimate last hurrah for the monolithic era.

Now, I just want to see NEAR embrace rollups, DAS & become a great DA layer!

However, the usual caveats apply: NEAR's token distribution is highly centralized, the validator set is limited, and in its current state the 4 shards are more like 4 subnets, i.e. they do not share security. So, yes, you're trading off security & decentralization short-term.