You may have heard “the more activity there is, the cheaper fees on rollups”, “modular architectures invert the trilemma” etc. Here, I just want to write a brief post about how scaling actually works.
Users use rollups and pay a consolidated transaction fee. This is broadly an aggregate of three components, corresponding to each layer:
Rollup fees: for operating the rollup
Settlement fees: proof verification & state root commitments to the settlement layer
Data fees: compressed transaction data to the DA layer
Rollup fees actually work like monolithic blockchains. Until the rollup’s limits are hit, they can be negligible. However, if the rollup is saturated, the rollup fees can start creeping up, same as monolithic chains. [Do note that for equivalent rollup vs. monolithic, the rollup will have much higher limits.] There are ways around this — by having sharded rollups with recursive proofs etc. Or simply using a different, less busy rollup.
Settlement fees is where the “more active -> lower fees” comes in. Today, early ZKRs have limited activity, so this is often the dominant component, and largely why it’s often mentioned. This is a fixed or poly-log fee per batch, so the more transactions there are, the lower the fee per transaction is. As activity ramps up, this tends towards negligible. I’d estimate anything over 100 TPS and this is close to zero. For ORs, this threshold is much lower, <10 TPS.
Data fees: this increases roughly linearly with transactions (ORs) or users (ZKRs). With EIP-4488 this goes down by 5.3x, and for early ORs compression can reduce it by another 10x (so, 50x total). ZKRs already do compression well. With a dedicated DA layer with data shards and data availability sampling, the more decentralized the network is, the more data availability capacity there will be, leading to higher throughput for rollups. So, this is where the inversion of trilemma comes into play.
Long term, I expect both settlement fees & data availability fees to tend towards negligible, as there’s near limitless capacity available. So, the rollup fees may end up being the dominant component. In that case, the efficiency of each rollup will be the differentiating factor. Luckily, we have smart solutions like recursive ZKPs & ASIC provers being researched that can collapse this down to negligible too.