Rollup-centric Ethereum roadmap: November 2021 update
Overwhelming demand for the Ethereum network combined with by-design constrained supply has in recent months led to skyrocketing gas fees. This has had a knock-on effect with rollups also seeing significant increases. Currently, AMM swaps cost ~$5 on optimistic rollups and ~$1 on zk rollups — which is too damn high. Do note that these are still very early beta & unoptimized rollups. Neither Optimistic Ethereum nor Arbitrum One have implemented data compression. With compression, we could see these fees go down by 10x. ZK rollups do have very efficient compression implemented, but early rollups have a different issue — not enough activity. The good news is as activity goes up, the transaction fees on zkRs will decrease significantly — especially STARK rollups. But optimizations and building activity will take time, and even then, it’s not enough.
Back to Ethereum, the long-term solution has always been data sharding, but with the community and developers opting to prioritize The Merge instead, has been pushed back to late 2023. We need shorter-term solutions. Vitalik details an update to how we can unlock as much data availability for rollups as quickly as possible. For details, please read that. Here, I’ll just state my quick (PS: lol, maybe it’s not so quick after all) opinion & speculation on the matter.
With rollups, especially ZKRs, the whole “TPS” thing is irrelevant. But for illustrative purposes, I’ll add what the average TPS at each step would be for a ERC20 transaction. For dYdX transactions, multiply this number by 3. (Yet another point of evidence that TPS is useless — one would have thought highly complex derivative trades with cross margin, oracle updates multiple times a second etc. would cost less than a simple ERC20 transfer.)
Step 1: EIP-4488/90
You can read about my thoughts on EIP-4488 here. Since then, we also have EIP-4490, which is a simpler alternative. These have broad community support, and the timeline is ASAP. On Friday’s Core Devs call, both will be discussed. EIP-4488 is the preferred solution, but a little more complex, so client implementers will have to decide if it will impact The Merge timelines. If it turns out that EIP-4488 will delay The Merge at all, the alternative is EIP-4490, which is a one-line change. Let’s wait and see, but I’m optimistic one of these will happen pre-Merge. As for timelines, we’ll also find out tomorrow. My best guess would be Jan/Feb 2022.
EIP-4488 will decrease calldata costs by 5.33x (EIP-4490 is 2.66x), though throughput only sees a minor bump to 5,000 TPS. How much this will decrease fees by is a complex matter (see my post above), but at constant demand, we should expect ~5x for optimistic rollups.
Step 1.5: Optimized rollups
This is not part of the Ethereum roadmap, but more about the rollups side. Still, it’s crucial information. Through the course of 2022, I’d expect rollups to continue developing. Arbitrum Nitro will introduce the first implementation of calldata compression. No timelines are given, but I’d speculate Nitro is coming early 2022. Optimism is also working on compression. I’d expect both to continue iterating, and delivering mature compression by the end of 2022. As mentioned above, this can lead to a 10x further decrease in cost over EIP-4488. So, we’re looking at a 50x reduction in a year’s time.
With ZKRs, things are a little more complicated — it totally depends on how much activity there is. If we see a ZKR take off in a big way, the verification costs will essentially be amortized to negligible, and the calldata costs will dominate. So, your dYdX transaction will cost only 16.1 gas, and the baseline ERC20 transaction 48 gas. 10x is definitely possible — especially for STARK rollups, so once again, we’re at 50x from today.
Step 2: Few data shards
Instead of implementing to full data sharding spec, we’ll first start off with a fewer number of shards, e.g. 4 shards. As a side note, I’ve talked about this off and on in casual comments, and wrote a short post about it.
With 4 shards, in addition to EIP-4488/90, we’re now looking at ~10,000 TPS. As for cost, we’ll see dedicated fee markets on data shards starting from zero, and I expect transaction fees to more than halve. It’s unclear to me how the execution layer’s calldata market will work in tandem with the new shards, though. Speculation on timelines: it’s implied to be similar in scope to Altair. Given that, I’d say early 2023 is a reasonable target.
Step 3: 64 data shards
This is the good old data shards v1 spec as we have come to know and love! We’ll see capacity increase all the way to 85,000 TPS, or 250,000 TPS for dYdX type transactions. This is where almost all rollup calldata is settled on data shards with dedicated fee markets, and I’d expect transaction fees to absolutely plummet. It’s hard to say by how much, so let’s take a conservative 8.5x (to go with capacity).
When does this happen? Again, totally speculating here: late 2023 is possible, but conservatively, it could be early 2024 due to Step 2 coming first.
This means, at constant demand, we can expect transaction fees on rollups to plummet by over 1,000x from the status quo on rollups today. But, of course, this is a very naïve illustration. It doesn’t mean that fees are going to be $0.0001 or something — of course there’ll be massive demand induced by these lower fees. On the flip side, a lot of the overwhelming demand for Ethereum is due to speculative activity in a bull market, which will almost certainly vanish in a bear market. Indeed, just 5 months ago, gas price was 10 gwei, and swaps even on unoptimized rollups were $0.30 or so. So, it’s really hard to say where thing settle. But the important thing to know is that we’ll have massive capacity with very low fees on rollups in a couple of years.
Step 4: Data availability sampling
DAS is a magical solution that lets you verify data availability with only a fraction of the data. So, to verify a 1 MB shard block, you only need to download a few kBs! This greatly increases security to the point that even a 51% attack is insufficient. Expect DAS to roll out through 2024 in stages. After this step, sharding is done!
(For those wondering — what happened to “Ethereum 2.0” execution shards? My speculation is those will never happen, and Ethereum shards will be data-only. Rollups & data sharding in tandem are simply a far superior solution than execution sharding. Instead, the Ethereum execution layer will head straight to zkEVM sometime mid-2020s, and then, if required, we can have zkEVM-shards late-2020s. Totally speculating here, though. I know some still want to make execution shards happen.)
Speculative steps: Expanding data shards
This is obviously much more speculative, and not part of Vitalik’s post. After DAS, sharding is done. But, just like Ethereum has increased its gas limits incrementally, we can expect each shard’s capacity to increase over time as bandwidth improves. According to Nielsen’s Law, we should expect 50x bandwidth — I don’t quite buy that, but the point is there are massive gains to be had over time. Additionally, as the networking layer matures, as we have more validators, and it gets cheaper to run the Beacon Chain (ZK-Beacon Chain, anyone!?), we can also add more shards. As we have speculated before, we could have dozens of millions of TPS by the end of the decade, and this does not even account for various new breakthroughs.
Elephant in the room: volitions
But, of course, the beauty of the modular architecture means that ZKRs need not wait for Ethereum’s roadmap to unfold. They can simply use alternative DA solutions — at a trade-off to security, of course. Decentralized validium options are still more secure than sidechains and alt-L1s. So, zkSync 2.0 will have zkPorter in early 2022. StarkNet will also have a range of DA options, including permissionless & decentralized solutions unlike the current StarkEx DAC. The volition system for StarkNet will be introduced in January 2022, though we don’t know when the first in this “range of DA options” will roll out — probably later in Q1 2022.
There’s a lot more in Vitalik’s blog post, including how expired history will be handled in a data sharded world. Highly recommend it! I’m more excited than ever for Ethereum’s massively ambitious rollup-centric roadmap — as I’ve said many times before, in collaboration with rollups and alt-DA layers, this is the ONLY WAY the blockchain industry scales to global ubiquity. However, it’s worth remembering that the transition to rollup-centric Ethereum remains a years-long journey. While that may seem like a long time, remember that this is the absolute bleeding edge of blockchain tech, and in the new paradigm, we’re still early. We’re now at the same point with rollups & data shards where Bitcoin was in 2009 and Ethereum was in 2015. Enjoy the ride!