Since EIP-4844 was implemented, more than a dozen rollups have started using blobs to post their batched transaction data to L1. In this article, I will
Explain my methodology for analysing rollup cost data
Measure the cost reduction rollups experienced thanks to EIP-4844
Explore how the blob fee market has evolved following Dencun (the Ethereum upgrade which activated EIP-4844)
Investigate the impact of EIP-4844 on the user experience of L2s
Dataset
For this analysis, I used an open-source Dune dataset called rollup economics. I created this dataset in July 2023, and have been working to expand it for the past year, assisted by contributions from Michael Silberling from OP Labs, Lorenz Lehmann from GrowThePie and Landon Gingerich from Matter Labs.
That dataset currently has 4 tables in it:
L2 revenue
- This table captures the revenue earned by rollups. It aggregates the gas fees paid by users each day.
L1 data fee
- This table captures the gas fees paid by rollups to post their transaction data to L1. It collects all the batch posting transactions made by rollups and calculates the gas fee paid, amount of data posted, gas used .etc
L1 verification fee
- This table captures the gas fees paid by rollups to verify their proofs on L1. It collects all the proof verification transactions made by L2 and calculates the gas fee paid, size of proof and more
L1 blob fee
—When rollups post transaction data using EIP-4844 blobs, a blob fee is charged separately from the normal gas fee. This table collects all the batch posting transactions and calculates the blob fee.
L1 fee
- Aggregates the daily data fee, verification fee and blob fee into one table.
How has EIP-4844 impacted the costs of L2s
As the table below shows, most of the rollups in the dataset experienced a 90%+ reduction in their L1 costs after adopting blobs.
Source Code: dune.com/queries/3811775
The biggest expense that these rollups faced prior to EIP-4844 was the cost of storing their batches in the CALLDATA of L1 transactions. Now, they store their batch data in blobs, which are much cheaper because they are temporary. Blobs are pruned after two weeks, allowing them to be priced cheaper than CALLDATA, which is stored in history forever.
As the chart below shows, rollups' weekly total L1 costs dropped 97% between the week before Dencun and the first week of June, from $19,023,381 to $482,181.
Source Code: dune.com/queries/3815141/6416367
It took a few weeks for the full impact of EIP-4844 to be felt because not all rollups upgraded to enable blobs at the same time. For example, the OP chains started using blobs on March 14th, three days after Dencun. But Scroll did not start using blobs until April 29th.
How has the blob fee market evolved?
When blob-enabled rollups post their batch data to L1 they pay both a normal gas fee and a separate blob fee for the transaction.
As we can see in the chart below, rollups spent a fraction of a dollar on blob fees in the first two weeks after Dencun. Then, the total daily spending on blob fees shot up on March 27th and climbed to an all-time high of ~$290,000 on April 2nd. After April 3rd, the spending on blob fees crashed back down to a fraction of a dollar and has stayed down ever since.
Source Code: https://dune.com/queries/3815223/6416515
Again, it is important to note that blob fees are only one part of the cost of posting batches.
What caused that huge spike in blob fees?!
Blobscriptions activity was what pushed the blob base fee to crazy heights in late May and early April. Blobscriptions was a project where people embedded text and image data in blobs to create pseudo-tokens they could trade on the Ethscriptions market.
Then, when people stopped making Blobscriptions, the blob fee market became cheap again. Right now, the blob base fee is 1 wei. This means one unit of blob gas costs about $0.0000000005. For context, at the top of the Blobscriptions craze, one unit of blob gas cost $67.
Source: https://dune.com/hildobby/blobs
Credit to Archwizard Hildobby
Currently, not enough blobs are being submitted in each block to hit the blob "target," which would activate the blob fee market and lead to the blob base fee rising and falling with demand.
Source: https://dune.com/hildobby/blobs
Credit to Hildobby again (he's the best!)
How did this affect the L2 user experience?
After the Dencun upgrade, rollups’ costs dropped by an order of magnitude, so they could afford to reduce the amount they charged users in gas fees.
Source Code: https://dune.com/queries/3817983/6421319
Source Code: https://dune.com/queries/3818798/6422780
Source Code: https://dune.com/queries/3818853/6422859
As the charts above show, the gas fees for making common transactions on rollups, such as ERC20 transfers and DEX trades, have fallen dramatically since Dencun. Today, making an ETH transfer costs less than 1 cent on rollups like Optimism and Arbitrum.
What's next?
EIP-4844 has successfully rollups’ onchain costs, thereby enabling cheap transactions for rollup users.
I anticipate that, as more rollups launch and start posting blobs, the blob "target" will be hit more often, and blob fees will go up. But even when the fee market is active, it's still dramatically cheaper for L2s to submit their batched data to L1 as blobs than to use the CALLDATA of a regular transaction.
I look forward to seeing all the new use cases that this low-fee environment enables!
Thank you for the article! I have a question
"The biggest expense that these rollups faced prior to EIP-4844 was the cost of storing their batches in the CALLDATA of L1 transactions. Now, they store their batch data in blobs, which are much cheaper because they are temporary". Is it possible to estimate roughly which part of total fees relates to storing data in blobs now. I assume in the past (before EIP-4844) it was the vast majority, but what about now? What is the main "spender" of fees for rollups?
If it is not DA anymore will appchains consider blobs as a DA instead of using other DAs like Celestia/Avail/etc?
You made it a lot easier for me to understand the technicalities behind Blobs, EIP-4188 with the touch of Onchain Data.
I'm finding your research quite helpful...