Why the transition to rollup-centric Ethereum is a years-long journey
I’ve mentioned in multiple comments and posts that I expect rollups to mature “in a couple of years”, so around late 2023 / early 2024. In this post, I’ll go through the roadmap and expected evolution of rollups, and why we’re not going to see adoption overnight. I’ll assume that you are familiar with how rollups, the role and risk of sequencers etc. Do note that there are many unknowns, but I’ve estimated things to the best of my knowledge.
Application-specific rollups (2020)
The journey to rollups began with application-specific rollups, starting with Loopring in March 2020. zkSync and DeversiFi (validium, not a rollup) went live in June 2020. Of course, we have over a dozen application-specific rollups today, with dYdX being a runaway success this month. You’ll note that most application-specific rollups are zkRs, and really, the journey began with EIP-1679 in the late 2019 Istanbul upgrade.
Smart contract rollups (2021)
Technically, the first smart contract rollup went live in January 2021 with Optimistic Ethereum, albeit with a strict whitelist. Since then, we’ve seen Uniswap V3, Kwenta, Chainlink and 1inch roll out, with many more projects being deployed over the next couple of months. Of course, Arbitrum One will be the first public launch with hundreds of projects deployed in just a few days’ time.
But these early application-specific and smart contract rollups come with severe limitations:
- Whitelists, as mentioned above, though Arbitrum One removes them in a few days.
- Throughput limits
- Upgradable L1 smart contracts
- Centralized sequencers
- Permissioned provers
- Unoptimized smart contracts, missing compression and aggregation techniques
- Data availability bottlenecked by the Ethereum execution layer
- Sub-optimal EVM, lacking precompiles
This is the list of reasons why it’s going to take a couple of years for rollups to attain their final form. So, let’s run through them.
Decentralizing L1 smart contracts (2021/22)
Aside from bugs, L1 smart contracts are the biggest security risk for most rollups today. Most early rollups are centralized, and in most cases, the smart contract’s multi-sig signers can steal your funds. It’s quite understandable why this is the case — rollups are bleeding-edge tech, and the ability to upgrade is crucial. But you must understand this risk before you ape in to a rollup.
Some rollups are less centralized than others in this respect. For example, zkSync 1.x does have upgradable contracts, but it’s a N of 15 timelocked multi-sig by various reputable members of the Ethereum community, where N dictates the timelock. Based on how critical the upgrade is, there’s a minimum timelock of 3 days for 12 of 15, but usually 2 weeks. Personally, I’m OK with this setup, as even if all a majority of highly reputable people are compromised, I still have time to exit the rollup.
Decentralizing L1 smart contracts is very much about progressive decentralization. We can start with centralized upgrades, move to multi-sig, then timelocked multi-sig. From here, we can have governance tokens enforcing timelocked upgrades. The end goal, though, is immutable smart contracts, like Uniswap. Even here, it’ll be a progression where different smart contracts can become immutable at different times. When the bridge smart contracts are made immutable, it’ll be the turning point. I believe Arbitrum are trying to do this by the end of the year, though I expect most rollups to decentralize their L1 smart contracts in 2022.
Decentralizing sequencers & provers (2022/23)
Note that Hermez is first to decentralize its sequencers and provers in late 2021 with a sequencer auction mechanism. StarkNet is scheduled to be decentralized in mid 2022. All other rollups have committed to decentralizing their sequencers and provers, but haven’t committed to dates yet. Different rollups have different models, of course, and may decentralize different aspects at different times. For example, Optimistic Ethereum will likely have permissionless fraud provers (i.e. anyone can submit fraud proofs) before sequencers are decentralized.
Some rollups may never decentralize their sequencers, for maximum efficiency. Particularly for application-specific rollups, this makes sense. I’ll remind you again that this is not a security risk.
Smart contract and rollup optimizations (2022 onwards)
At this time, most smart contracts being deployed are dumb replicas of their L1 counterparts with minimal changes. Likewise, rollups themselves haven’t leaned into compression techniques for calldata or using signature aggregation. As a result, we’re seeing “only” 90%-95% reduction in fees. As smart contracts optimize for rollups, and rollups evolve, we should expect to see a further order of magnitude reduction in fees.
In addition, rollups will continue to increase their throughput, and implement advanced state size management techniques like state expiry. All of this will lead to increasing throughputs and potentially lower fees.
All of this will, of course, be a gradual evolution over time, and I fully expect by the end of 2022 both rollups and smart contracts on them to be far more optimized than now. Fee reductions will be 99%, and we’ll be bottlenecked by the limits of L1’s data availability.
Data shards (2023)
Speaking of, data shards are when the floodgates open for rollups. I’d estimate the most likely release for data sharding will be early 2023. This is when rollups will be able to do tens of thousands of TPS. Over the years, more shards will be added, up to a current maximum of 1,024; and over the years as bandwidth and storage improves, we’ll see each shard expand as well. Long-term, rollups are all set to do millions of TPS over Ethereum data shards.
I fully expect 2023 to be the year where a vast majority of transactions in the blockchain industry happen on zkRollups. And yes, I expect optimistic rollups to make the transition to validity proofs in 2023.
L1 VM upgrades (2023 onwards)
I don’t understand this well, so I won’t go into details. The EVM is unfriendly to rollups, especially zkRs. This is understandable — L1 EVM has a burden of hundreds of billions of dollars and is relatively ossified as a result, and cutting-edge precompiles and elliptic curves are very high-risk. Fortunately, rollup developers are brimming with ingenuity and have conjured effective workarounds anyway. But for rollups to attain maximum efficiency, we’re going to need VM upgrades on L1. These will happen, but probably after data shards and state expiry, so realistically late 2023. One showerthought I have is to do a rollup-centric execution shard with a new, custom VM designed purely for zk proofs, rather than grafting the EVM.
Make no mistake, the future of the blockchain industry is rollups + data shards. There’s no other solution known that can scale to millions of TPS in a highly secure, decentralized, trustless, permissionless and credibly neutral manner.
But it’s going to be a bumpy ride, there’ll be a metric ton of FUD from L1-maxis, but we’ll get there over the next couple of years.