Common misconceptions around rollups melt away when you make a simple perspective shift: anything any L1 can do, as a rollup it can do it better.
L1 does X TPS -> Rollup does 10X TPS
L1 does X finality -> Rollup does 0.1X finality
L1 does 3 programming languages -> Rollup 10
L1 does 3 upgrades / yr -> Rollup does 12
L1 has novel fee model -> Rollup has multiple novel fee models
L1 does yearly state expiry -> Rollup monthly
3 L1s suck at interoperability and fragment liquidity -> 10 rollups have superior interop, can even share liquidity (see: dAMM)
All above for a simple reason: a rollup only needs to provide ephemeral censorship resistance & liveness for a few minutes; while an L1 needs CR, liveness & safety forever. By cooperating with settlement & DA layers, frees up a rollup to be infinitely more innovative & efficient.
Caveat: Of course, rollups are early. But we have proven the concept works brilliantly. dYdX & Immutable X have proven zero gas fees are possible at scale. Now, it's all about iteration, and adding features like decentralized sequencing to exceed anything any L1 can ever do.
Of course, none of this would be news to anyone who has read my posts. For a bit more detail, see here: polynya.medium.com/addressing-com…
Note: you can also extend this to settlement & DA layers. A specialized settlement layer can be more secure, decentralized and sounder money than an L1 that tries to do it all. A DA layer with DAS can be infinitely more scalable - higher decentralization -> higher throughput etc.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
With today's technology, it would take over 10,000 years for a crew to travel to Proxima Centauri. In 50 years, we could have the tech to do it in less than 100 years. So, why not wait? The ye olde crew can milk a lot of attention in the short term, and be absorbed later.
This is the fate that awaits megalomaniac blockchains (aka L1s). In a couple of years for now, they will be utterly obsolete, leapfrogged >1,000x, but there's still users & devs to onboard now while the next-gen tech (rollups/DA layers) is maturing. Soon, they'll be assimilated.
So, please, feel free to use & develop for megalomaniac blockchains today, but be aware they'll be heavily crippled and pretty useless tomorrow. Vastly prefer projects that have a clear path to a modular future. The ones with rollup-deniers are NGMI - stay away from them.
One of the most underappreciated benefit of rollups vs high-TPS L1s is efficiency of liveness, censorship resistance and finality. In the latter, the consensus protocols have to offer permanent CR, L, and safety. In rollups, you only need ephemeral CR, L for a few minutes.
1/8
ORs generally settle every couple of minutes, SNARK rollups like Loopring ~12 minutes, STARK rollups like dYdX ~1 hour. It could be longer if they are less active, but as activity ramps up and the tech matures, I'd expect these times to decrease rapidly, eventually few sec.
2/8
Rollups that want to maximize ephemeral CR & L, can run proof-of-stake consensus just like their MC ancestors. However, they only need to do this for a few minutes instead of forever! As a result, they can be far more efficient. Which also enables more rapid finality.
3/8
Web3 is a positive sum game. Solana's technology has abysmal scalability, capable of addressing <1% of TAM long term. The notion of a single database addressing much of the global audience is ludicrous in web2, and a single ledger doing the same is no different with web3.
I welcome more centralized solutions like Solana or Polygon PoS to onboard the surplus demand in the short term here & now. However, Polygon has a much more ambitious and future-leaning roadmap than Solana with ZKRs, scaling well beyond what Solana can ever dream of.
2/6
Solana's bizdev and marketing teams have done a phenomenal job, and the audience they onboard till StarkNet, zkSync 2.0 and Polygon Hermez/Miden mature will be well earned. But this is <<1% of where web3 is headed long term. Remember, a single ZKR can outscale Solana.
3/6
General CR is same as settlement layer - you can exit from there.
Ephemeral CR (minutes, potentially seconds in the future) is as decentralized as sequencers are. Can even run PoS consensus with many validators for max CR, at the cost of efficiency.
I imagine most ZKRs will take the middle ground, by decentralizing their sequencers and provers just enough for the ephemeral CR. Just needs CR for a few minutes till verified on settlement layer. Nothing stops a ZKR from going all out with PoS BFT CR, but this is v. inefficient.
However, on the other hand, some may choose to keep their sequencers centralized in specific cases for max efficiency & DDoS resistance. There's no real incentive for a centralized sequencer to censor - they'll just lose business to competitors.
Understandably, the rhetoric is still entirely monolithic blockchain (L1) centric, presumably a reflection of people's bags - they don't want progress.
The real question is which projects are best specialized in these categories?
- Execution
- Settlement
- Data availability
I generally stick to talking about the specialized execution layers, i.e. rollups/validiums/volitions, but the truth is all three are interlinked and as long as people continue to have monolithic-centric biases the discussion is stifled. Let's start thinking in E, S and DA terms.
Example of why monolithic/L1-centric thinking is plain wrong. E, S and DA can be different. The correct comparisons should be E vs E, S vs S, DA vs DA; not L1 vs L1.
The magic of validity rollups: the more activity there is, cheaper it gets. Some quick estimates:
~12,000 TPS dYdX-type transactions on Ethereum possible today.
~250,000 TPS with V1 data shards (~2023).
4 million TPS with optimized data shards.
~100 million TPS by ~2030.
Yes - that's dYdX-type complex derivative trade transactions on validity rollups settling on Ethereum alone. These are v. conservative worst-case estimates not considering inevitable improvements in tech, further breakthroughs, data compression optimizations, non-Ethereum DA etc.
Circling back to dYdX - proof verification is still ~80% of the cost, so there's enormous headroom for growth and costs will plummet as activity rises. Eventually, each batch can be a million transactions & verif. cost will be negligible. We're at the very peak of the iceberg.