Translation: Blockchain in Vernacular

In this post, we aim to present a statistical overview of the current state of L2. We will explore the significance of the reduction in L2 transaction fees following the Dencun upgrade in March, examine the evolution of activity on these networks, and highlight new challenges arising from MEV activity. Additionally, we discuss potential obstacles to developing MEV tools and solutions on L2.


1. Positive impact of Dencun upgrade: L2 adoption

1) Transaction fees dropped 10 times

Ethereum Layer 2 (L2) transaction fees consist of two parts: the cost of executing transactions on L2, and the cost of submitting batches of transactions to Ethereum Layer 1 (L1). The specific transaction fee structure and sorting rules of different L2s vary depending on their development stage and design choices.

For example, Arbitrum uses a first-come, first-served basis (FCFS), where transactions are processed in the order they are received. In contrast, Optimism (OP Mainnet) and Base (both part of OP Stack) use a Priority Transaction Fee Auction (PGA) model that includes L2 base fees and priority fees. Users can choose to pay a higher priority fee in order to be included in the block faster and earlier. Understanding these fee structures is critical to understanding the growth of the ecosystem and the MEV dynamics.

Historically, Ethereum L1 fees made up the majority of total fees when users transacted using L2, accounting for over 80%, as shown by the black bars in the figure below. However, since the Dencun upgrade on March 14, L2 has switched from using calldata to a more cost-effective method of submitting batches to L1 called "blobs 1." This temporary storage method includes its own fee auction mechanism, including a blob base fee and priority fees.

Since the Dencun upgrade, L2 has paid a significant reduction in L1 fees - the chart shows that the transaction fee composition of the OP Stack chain has changed significantly, with L1 fees falling from 90% to just 1%, while L2 fees now account for 99% of the total fees. This shift has led to an overall decrease in the average total transaction fee of L2 by about ten times, for example, the average transaction fee on the OP mainnet has dropped from about $0.5 per transaction to $0.05.

2) L2 activity surges

After the cost dropped, L2 activity and usage increased significantly, as evidenced by the surge in L2 transaction fees shown in the chart above. Notably, on March 26, Base’s average transaction fees exceeded their pre-upgrade peak. To accommodate more transactions and reduce network congestion, Base raised its transaction fee target starting March 26 and has made multiple adjustments since then.

The chart below shows L2’s daily transaction volume, demonstrating the significant growth of networks such as Arbitrum, Base, and OP mainnet. Specifically, Base has quadrupled its daily transaction volume and now processes approximately 2 million transactions per day.

While it’s difficult to determine whether this is the result of natural growth or the impact of incentive programs and Sybil attack activity, there has been a noticeable increase in active addresses and decentralized exchange (DEX) volumes across all major L2s following the EIP-4844 upgrade, particularly on Base and Arbitrum.

3) Asset transfer to L2

With improving market conditions and the Memecoin craze triggered by WIF on Solana, the total locked value (TVL) of L2 has continued to rise since the end of last year. Notably, Base has become the fastest growing chain, and its total locked value recently surpassed the OP mainnet.

Since the beginning of March, Base has had about $1.5 billion in USDC inflows, part of which is Coinbase moving customer and corporate funds to Base. According to data provided by Artemis, Ethereum's capital outflow to major L2s has reached $14 billion since January 2024 through 11 major bridges. Arbitrum leads with about $7 billion, followed by zkSync, Base, and OP Mainnet. Further data shows that Debridge Finance (a bridge widely used in EVM chains and Solana) confirms that Arbitrum and Base are the main recipients of all capital outflows.

4) Bad news: Dark Forest is expanding amid cheaper transaction fees

When we examine transactions further, we find that bot activity is driving up L2 transaction fees and rollback rates. We explore this more fully in the next section through a case study using Base data, highlighting the impact of cheaper transaction fees on L2 after the Dencun upgrade.


2. Dencun’s upgraded L2: Like Ethereum before Flashbots without a memory pool

1) Network congestion

Challenges began to emerge: On March 26, Base’s average daily transaction fee experienced a short-term surge, even exceeding the level before the Dencun upgrade. By June 3, Base adjusted its transaction fee target from 2.5M Gas/s during the Dencun upgrade to 7.5M Gas/s, bringing the average transaction fee back to about 5 cents.

The contracts that consume the most transaction fees on Base include Telegram trading bots such as Sigma and Banana Gun, as well as wallets and decentralized exchanges such as Bitget and Uniswap. In addition, a large number of unlabeled contracts involve activities such as token minting, memecoin trading, and atomic arbitrage.

By comparing the behavior of popular Telegram bot routes like BananaGun, it is clear that their transactions generate significantly higher transaction fees compared to other transactions. After the upgrade, users of the BananaGun Telegram bot paid peak transaction fees of up to 30 Gwei to execute trades on Base. This rate is currently stable at about 3 Gwei, which is 43 times higher than the transaction fees paid by other transactions.

When analyzing the average monthly trading fees paid by all popular decentralized exchange (DEX) trading bots on Base, compared to all other non-Telegram bot trading (black bars), it is clear that trading bot users bear significantly higher transaction fees.

2) A surge in high rollback rates

Another important indicator of blockchain importance is the transaction rollback rate on the network, which we also observed an increase on %20L2s%20 after the %20Dencun%20 upgrade, especially on %20Base%20, Arbitrum%20, and %20OP%20 mainnets. (The rollback rate refers to the proportion of transactions on the blockchain that failed to be successfully confirmed for various reasons.)

Currently, Ethereum’s transaction rollback rate is about %202%, while Binance Smart Chain and %20Polygon%20’s rollback rate is about %205-6%. Before the upgrade, Base%20’s rollback rate was about %202%, but then rose to about %2015%, peaking at %2030%%%20 on April %20. Similarly, Arbitrum%20 and OP Mainnet have also seen periodic surges in failed transactions, ranging from %2010%%%20 to %2020%%%20.

After further analysis, we noticed that the high rollback rate on L2 does not necessarily reflect the experience of every ordinary user. Instead, these rollbacks are likely coming from MEV robots.

Using the following heuristic query, we identified a set of routing contracts with bot-style activity — contracts that appear to experience high rollback rates when executing MEV withdrawal transactions:

Since Dencun was upgraded,

  • Active Routing: The contract has processed more than 1,000 transactions. Limited Interaction EOA: Fewer than 10 EOA (Externally Owned Account) wallets have interacted as a transaction sender.

  • Sender distribution: Less than 50% of transaction senders sent only one transaction, indicating that this route is unlikely to be used by retail users.

  • Behavioral patterns: Transaction histories either cover exactly 24 hours or show multiple transactions within a single block, indicating non-human behavior.

  • Exchange concentration: More than 75% of successful transactions involve an exchange.

  • Detected MEV transactions: More than 10% of successful transactions utilized the atomic MEV strategy, detected based on hildobby's heuristic method 2.

Based on these criteria, we detected 51 routers that likely represent a conservative lower bound of bot activity on Base. We divided all transactions processed by routers on Base into two groups and then performed a comparative analysis between them. We found that the average rollback rate for transactions of router contracts with bot-like operations was 60%, while the rollback rate for other transactions was about 10%, and the rollback rate for bot-like operations was six times that of other transactions.

Based on the above data, we can conclude that bot activities, such as MEV bots and Telegram bots, are likely one of the main reasons for high transaction fees and rollback rates on Base.

L2's single sequencer infrastructure, coupled with the lack of a common memory pool, facilitated dominant MEV strategies involving massive sequencer abuse. These strategies significantly contribute to network congestion, especially for L2s like OP mainnet and Base that employ priority fee auctions (PGA). The result is not only network congestion, but also wasted block space due to rolled back transactions and transaction fees paid by MEV seekers. This situation mirrors the state of Ethereum before Flashbots, but unlike it, there is no sandwich attack situation on L2 due to the current lack of mempools.

3) How big is the MEV on L2?

Gaining insight into MEV activity on L2 is critical. However, to date, there is no consensus figure for L2 MEV verified by multiple sources and robust methods. In addition, real-time monitoring data similar to Ethereum (e.g. mev-inspect, libmev, eigenphi 2) is also lacking in terms of MEV transaction volume and seeker profits on L2.

So far, some datasets and studies on L2 MEV include:

  • Open source datasets built by hildobby on Dune Analytics (inspiration links: Sandwiched 1 | Sandwiches | Atomic Arb 3).

  • A research paper funded by Flashbots, Quantifying MEV On Layer 2 Networks 1, by Arthur Bagourd, Luca Georges Francois, quantified MEV on Polygon, OP Mainnet, and Arbitrum using the mev-inspect implementation.

  • The research paper, Rolling in the Shadows: Analyzing the Extraction of MEV Across Layer-2 Rollups 3, by Christof Ferreira Torres, Albin Mamuti, Ben Weintraub, Cristina Nita-Rotaru, and Shweta Shinde, quantifies the activity on L2 and discusses a novel MEV strategy that exploits the sequencer role and its L2 batch confirmation latency.

In addition to the above resources, Sorella Labs will soon release their MEV data indexing tool Brontes, which will be an open source repository available for Ethereum mainnet and L2. Flashbots and the Uniswap Foundation are seeking grants to expand the classification and quantification of L2 MEV. If you have done work in this area or are interested in collaborating, please contact the Flashbots market research team (@tesa_fb on Telegram)!

While further validation is still necessary, hildobby’s dataset on Dune Analytics serves as a valuable initial benchmark:

Over the past year, atomic arbitrage MEV volume on the six major L2 networks (Arbitrum, OP Mainnet, Base, Zora, Scroll, and zkSync) has reached over $3.6 billion, accounting for 1% to 6% of all DEX volume on each chain. This MEV volume is mainly concentrated in Arbitrum and OP Mainnet, but has recently shifted to Base and zkSync.

Compared to Ethereum, sandwich volume is significantly lower on L2, and atomic arbitrage volume is completely different, with a fourfold difference. This difference is due to L2's single sequencer setup, which by its nature does not introduce a mempool, so searchers will not be able to take advantage of sandwich MEVs observing user transactions from the mempool (unless there is a mempool leak or a sandwich from a single sequencer). Instead, strategies like atomic arbitrage, blind rollback, statistical arbitrage, and liquidation are the most viable options for searchers on L2.


3. Assessing the MEV market size: How much MEV revenue is left on L2?

Although it is difficult to accurately quantify the MEV market, we can refer to data from other ecosystems with MEV solutions for size comparison:

On Ethereum L1, annual validator revenue from MEV-boost blocks is approximately $968 million (estimated using an ETH price of $3,500); and the median value of a MEV-boost block is 4x higher than the value of a normal validator block.

On Solana, the additional MEV revenue collected by validators through Jito’s bundling service is approximately $3.38 billion (estimated using a SOL price of $130) based on an expected 5.2 million SOL per week.

While exact figures for Base’s MEV trading volume are not available, the market size can be estimated by analyzing the revenue of the Banana Gun Telegram Bot, one of the most active bots in this space. The bot’s trading volume on Base L2 is comparable to its trading volume on Solana, consistently generating over $100.2 million in trading volume per day, and therefore over $1.2 million in fees per day on each chain.

Please note that there may be significant differences in Banana Gun Bot's market share on Solana and Base. For example, Solana has several other important Telegram bots such as Sol Trading Bot and BonkBot, while there may be fewer Telegram bots that support Base. Therefore, Banana Gun's transaction volume does not calculate Base's total MEV revenue in proportion to their revenue on Solana.

However, consider another way to estimate this: In March alone, the Banana Gun Telegram Bot paid out over $23 million to Ethereum builders and validators! When comparing its transaction volume on different chains, its transaction volume on Base actually surpassed Ethereum during the week of March 26th and April 1st (as shown by the peak in the chart above), indicating that Base has significant MEV income potential.

Of course, there are significant differences between Base and Ethereum’s MEV ecosystem. Competition for MEV on Base is likely to be mild compared to Ethereum, meaning bots need to bid lower to validators. However, the Memecoin trading bot, which primarily operates through blind snapping and arbitrage, is still feasible within Base’s sequencer settings.


4. Summary

1) Calling for attention to MEV

Ethereum has built a complex MEV ecosystem where infrastructure tools serve participants at different levels of the supply chain. On a protocol level, MEV-boost allows validators to outsource the block building process through auctions. For seekers, bundling services similar to Jito Labs on Solana and FastLanes on Polygon, provided by Ethereum’s block builders, enable seekers to propose MEV strategies with rollback protection.

These services ensure that builders simulate transactions and only process those that will not be rolled back. In addition, private RPC services like Flashbots Protect provide retail users with a way to avoid public memory pools and avoid being sandwiched. The current form of L2 still requires considerable progress in developing a similar mature MEV infrastructure.

2) Why should we consider MEV solutions for L2?

Even without a memory pool, MEV still exists. MEV strategies such as statistical arbitrage (CEX-DEX arbitrage), atomic arbitrage (DEX-DEX arbitrage), and liquidations play a role in maintaining market efficiency, clearing out stale liquidity in AMMs and lending markets.

However, in the absence of mature MEV infrastructure like bundling services, negative externalities emerge. Without a memory pool, most MEV strategies default to spamming strategies, leading to:

  • Increased rollback rates in the network;

  • High gas fees, leading to network congestion.

By introducing bundling services and shifting the pressure of MEV competition from the chain to the sidechain, users can be protected from the high gas fees caused by MEV robot racing. Seekers can also get higher profits through rollback protection because the cost of failure can be reduced.

For L2s looking to adopt a shared sequencer, most solutions today require users to submit their transactions to a public mempool, reintroducing sandwich attacks. In this case, a private RPC like Flashbots Protect could provide protection to users by sending user transactions directly to block builders to prevent sandwich attacks, and even provide refunds of MEV or priority fees to provide users with better execution and better prices.

However, there are still open challenges for more complex MEV infrastructures:

  • First, the economics of search change as more value is paid to the sequencer, reducing the marginal profit of the searcher over time. This also raises the question of the sustainability of highly competitive search strategies in the long term. We expect market forces to be at play here, with common search strategies paying most but not all of the value to the sequencer, and uncommon search strategies paying less.

  • Furthermore, the order flow dynamics of existing MEV infrastructure like Ethereum’s block construction market are still evolving rapidly. At the time of writing, they have contributed significantly to the centralization of the block construction market and the rise of private memory pools on Ethereum L1. How to ensure a competitive and fair block construction market remains an open challenge.

  • Finally, MEV solutions for L2 may also differ from those currently available on Ethereum due to their faster block times, cheaper block space, and relatively more centralized governance. It is unclear whether fast block times, such as Arbitrum’s 250 ms blocks, are compatible with the current performance and requirements of existing MEV infrastructure. Moreover, the abundant and cheap block space provided by L2 changes the dynamics of search, making the spam problem more prominent and potentially requiring new solutions. More importantly, L2 is relatively centralized relative to other setups, such as Ethereum L1. In this case, it may be possible to impose additional requirements on MEV service providers, such as requiring block builders not to perform sandwich attacks on users, in order to achieve fair market outcomes.

As a final call to action for this post: we encourage researchers and teams to collaborate on defining novel L2 MEV strategies and designing L2 MEV solutions. Please share discussions under this post or refer to this post to apply for Flashbots research proposal funding. We are looking for funding to contribute to L2 MEV quantification and indexing tools and build a taxonomy of MEV states on L2. Please contact our market research team (@tesaspeech_balloon or @tesa_fb on Telegram) for more details on L2 MEV funding. As part of our efforts to explore L2 block construction on SUAVE, we are looking for people to join a community call to discuss blind rollbacks and their impact on sequencer design and throughput. If you would like to participate, please reach out to @dmarzrobot (@dmarzz) on Telegram.