Flashbots Research: How MEV Devours Blockchain Scalability Benefits

Written by: Robert Miller, Flashbots

Compiled by: Saoirse, Foresight News

Today, we put forward a new argument: MEV (Maximum Extractable Value) has become a major limiting factor in blockchain scalability.

As Ethereum and its Layer 2 networks, as well as mainstream public chains like Solana, compete to scale at the fastest speed, the economic constraints brought by MEV have already manifested across the industry. On-chain search behaviors are starting to occupy the majority of the main capacity of high-throughput blockchains in an astonishingly wasteful manner.

This is not a theoretical hypothesis or an isolated phenomenon. From Solana (where MEV bots consume 40% of block space) to the Ethereum Layer 2 ecosystem, this situation is ubiquitous. To quantify the impact, we conducted an in-depth analysis of the top OP-Stack Rollups supporting specific tracking endpoints, and the results revealed an industry-wide issue:

The garbage trading bots in multiple Rollups consume over 50% of the Gas, yet pay less than 10% of the fees.

From November 2024 to February 2025, the Base network will increase its gas processing capacity to 11 million gas per second, but almost all of it will be occupied by garbage bots (equivalent to the capacity of three Ethereum mainnets!).

The continuous demand for Gas by garbage robots has driven up user transaction fees.

The garbage trading market is highly concentrated, with over 80% of the garbage trading on Base dominated by two searchers.

Database sharding (such as Rollup), validity proof, optimization of databases or consensus mechanisms, and other technical scalability measures are certainly important, but relying solely on technology cannot solve the problem. While we have mastered the methods to build the throughput of basic technologies, the current market structure imposes economic constraints on scalability.

This article will analyze this market failure phenomenon, demonstrate its impact with data, and propose a new MEV auction mechanism aimed at addressing the issue.

Analysis of Trash Trading

To understand why block space is wasted, let's first break down a successful arbitrage trade:

Successful arbitrage trading example on Base

At first glance, this seems to be a model of efficiency: the search bot executes precise arbitrage, earning $0.12 and paying a fee of $0.02.

But the real cost of this successful arbitrage is shocking: for each successful arbitrage, the robot sends about 350 attempts to arbitrage (most of which fail). On average, a single successful arbitrage consumes about 132 million Gas—equivalent to nearly 4 complete Ethereum blocks. It should be noted that this is just one of the many robots participating in the competition, and the actual cost on the chain is even higher.

Now let's look at a typical failed attempt to understand the on-chain behavior of robots:

Examples of failed trades blindly seeking arbitrage opportunities

At first glance, this transaction appears normal: it executed successfully and did not involve any token transfers. The only clue is that it consumed about 2.6 million Gas (as shown in the image above).

In-depth tracking of its internal calls reveals that it initiates a series of calls to dozens of different DEX pools, querying the pool status through getReserves ( and slot0 ). These calls essentially retrieve asset prices on different DEXs.

Demonstrate the tracking example of repeatedly calling slot0 () and getReserves ().

The core logic of this robot is very simple:

Send transaction to the blockchain

Query prices from multiple DEX pools during execution.

Execute if there is an arbitrage opportunity

Terminate transaction if none

The above transaction reflects these four steps, ultimately terminating without executing any operations. In fact, it was just a high-intensity price query, consuming about 2.6 million Gas while merely reading the market status without any substantial action.

On public chains like Base, World, and Solana, this strategy has become the mainstream way to extract MEV. A few successful trades must pay the price for a large number of failed attempts, which is a rational choice for seekers but causes systemic inefficiency for the network.

A lot of resources are used to read prices without generating real value. And it's not just one searcher, all searchers have to employ this tactic in order to capture atomic-scale MEVs. The final result is as shown by the data: the public chain is blocked by spam transactions, and the transaction fee rises due to spam transactions. (Note: Atomic-level MEV emphasizes the extraction of value achieved in a single on-chain operation (such as a single transaction or within a single block), and is commonly used in scenarios such as arbitrage and front-running, which take advantage of blockchain immediacy and transaction order.) ）

The root cause of garbage trading

It is not a coincidence that high-throughput public chains are blocked by junk transactions; rather, it is a direct and "rational" response caused by structural defects in the market: if seekers want to profit by reading the latest state of the block, they must blindly initiate transactions within the same block.

The arbitrage robot analyzed above is a typical case. Off-chain queries can capture the status of the last confirmed block, but this lags behind the MEV opportunities that are being created by transactions in the current building block. In networks such as Base or Solana, the native mempool is private, which means that searchers have no way of knowing how a user's transaction is performing and what opportunities it creates until a block is published. The only way to discover and capture arbitrage space is to have your own transactions included in the same block immediately after a user's transaction. Once you wait for the next block, the opportunity is preempted.

The rampant phenomenon of on-chain searching stems from the interaction of the following factors:

1. Transaction Expressiveness

Unlike traditional finance where traders submit simple static orders (e.g., "buy at price X"), seekers can create transactions as on-chain programs, embedding conditional logic based on the real-time market state to achieve complex responsive strategies that were previously impossible.

2. Switch to Private Memory Pool

To protect users from front-running, most high-throughput public chains set their memory pools to private. While this effectively defends against front-running, it also prevents seekers from seeing user order flows. Since seekers cannot react before transactions are on-chain, they can only initiate high-expressiveness transactions to blindly probe for opportunities on the chain.

3. Low transaction fees

The low-cost block space further amplifies on-chain search behavior. Searchers are well aware that the profit from a single successful arbitrage can cover the costs of numerous failed transactions, which is why they dare to send massive speculative trades to each block. The lower the Gas fee, the more complex logic searchers can write, pursuing more intricate strategies. (

4. Lack of an efficient auction mechanism

The competition among searchers lacks a formal mechanism for expressing trading order preferences. Since it is not possible to directly bid for the sorting of specific transactions within a block, the competition degenerates into a wasteful alternative: consuming more Gas. The main way searchers increase their chances of success is by consuming Gas in more positions within the block to increase the probability of transactions landing in the "correct position."

These four major factors have jointly given rise to "garbage trading auctions," an extremely wasteful mechanism that not only exacerbates network congestion but also fails to effectively capture MEV value. To quantify the inefficiency scale caused by garbage trading, we conducted data validation.

Research has found

Analysis confirms that MEV-driven garbage trading poses economic constraints on scalability.

We define junk transactions by identifying transactions that "query DEX repeatedly but do not transfer tokens." This heuristic method aims to locate systematic wasteful "backrunning" arbitrage behavior that could have been completed off-chain but is forced on-chain. We have implemented this method in both Python tools and the Dune dashboard, with detailed methodology provided in the appendix.

Due to the reliance of garbage transaction detection tools on specific RPC methods, current data analysis is limited to OP-Stack Rollup. However, data from the Ghost Logs team indicates that similar phenomena also exist on Solana, and other Ethereum Rollups (such as ZKsync and Arbitrum) have also been found to exhibit signs of garbage transactions.

1. Junk trading has systemic and universal characteristics.

First of all, this issue is systemic and widespread. Analysis of the OP-Stack Rollup indicates that spam transactions are not an isolated phenomenon, but rather a dominant force within the entire ecosystem. On chains such as Unichain, Base, and the OP mainnet, spam transactions typically consume more than 50% of the total Gas. This shows that it is a structural consequence of the current market design, rather than a localized anomaly.

2. The Gas consumed by garbage trades far exceeds the fees paid.

The second finding shows that, from the perspective of the chain, the efficiency of junk transactions is extremely low.

In all the Rollups we analyzed, there is a significant gap between the resources consumed by spam transactions and the revenue they generate. Compared to other users, spam trading bots consume several times the amount of Gas relative to the fees they pay. For example, spam bots on the OP mainnet consumed about 57% of the Gas but only paid about 9% of the fees, resulting in a gap of as much as 6 times.

The gap between transaction fee payments and gas consumption indicates that spam transactions impose significant external costs on the network while providing almost no corresponding value, which is a typical feature of a systematically inefficient market. This includes real waste of computing resources, as every full node is forced to execute these transactions, thereby raising the hardware requirements for all network participants.

In addition, we also analyzed how spam transactions in L2 affect the use of Rollup for L1 Data Availability.

Data shows that in one million blocks in February 2025, garbage bots on Base contributed approximately 56% of Gas consumption, 26% of L1 DA (Data Availability) usage, and 14% of on-chain transaction fees. The initial surprise regarding the proportion of DA usage by garbage bots was soon clarified as it relates to their transaction volume (rather than Gas consumption). This is reasonable because DA usage depends on data compression efficiency, not on Gas consumption.

3. Garbage trading restrictions offset the benefits of scaling.

Third, this inefficiency directly offsets the benefits of scaling. To measure the negative impact of junk transactions, we introduced a new metric: effective Gas throughput, which is the amount of user-available Gas processed per second after Rollup deducts the consumption by junk bots.

The trend is particularly pronounced at Base: in November 2024, the total gas throughput will be 15 million gas/s, while the effective gas throughput for users will only be 12 million gas/s. Over the next four months, the total throughput increased by 11 million gas/s, but the effective throughput remained virtually the same. In other words, almost all of the additional processing power is taken up by spam transactions.

Interestingly, after the end of February, the effective throughput began to align more closely with the growth trend of the total throughput. This seems to be related to market trading volume (and the resulting MEV): after the "Libra scandal" broke on February 14, the effective throughput started to grow again as the trading volume of Memecoins via Telegram bot transactions declined.

4. The continuous demand for garbage trading has driven up user transaction fees.

Perhaps the most direct impact on users is that the continued existence of garbage trading artificially raises the baseline for transaction fees, keeping them high for a long time.

Although the scaling efforts of Rollup have reduced nominal fees to extremely low levels (for example, around $0.01), making many natural users insensitive to the price, theoretically, if block space is sufficient and users are insensitive to the price, combined with the effect of the EIP-1559 fee market mechanism, fees should tend towards an absolute minimum. The vision of scaling is to create enough capacity to make this near-zero fee state the norm.

But the reality is not so. Seekers trying to capture MEV through junk trading are filling blocks with massive transactions, consuming a large amount of Gas. This behavior drives up block utilization, leading to a continuous increase in base fees, which reflects more of the systemic inefficiency of the MEV market rather than the genuine demand of natural users.

Although the fees borne by end users are still low, the overall level is much higher than what is actually needed. The crux of the matter is that innovative use cases that rely on large amounts of cheap block space, such as on-chain social networks or automated micropayments, are being excluded from the market as a result.

5. The garbage trading market is highly centralized.

Finally, the analysis shows that the searcher market for MEV junk trades exhibits extreme centralization characteristics.

To verify this, we counted which smart contracts consumed the most Gas classified as "garbage transactions" between block heights 26000000 and 26900000. At first glance, the market seems to have a high concentration at the top but is structurally dispersed.

But this appearance is deceptive. On-chain analysis shows that a common strategy used by seekers is to rotate the smart contracts used to send junk transactions, while consolidating profits into a fixed "profit address." By tracking the ETH transfer paths of successful arbitrage trades, we attempt to identify smart contracts controlled by the same operator. Although not all bots adopt this model, the leading bots generally do.

When the data is grouped by the profit address, the market concentration becomes extremely significant:

The results are clear, with just two institutions dominating more than 80% of the spam transactions on Base. This extreme concentration indicates that there are clear barriers to entry and that the current "junk trade auction" is not a truly competitive market. The lack of competition further weakens the price discovery mechanism, resulting in the public chain not only being unable to capture the true value of the extracted MEV, but also having to endure the negative externalities caused by spam transactions.

The Path Forward

We believe that blockchain should maximize valuable economic activities within limited block space.

From this standard, the current "garbage trading auction" mechanism is extremely inefficient: completing two arbitrage exchanges on Uniswap v3 only requires about 200,000 Gas, while achieving the same economic result on Base consumes about 130 million Gas. The efficiency gap is as high as 650 times, and narrowing this gap is the key to unlocking the true potential of scalability.

To solve this problem, we must first return to the four main reasons why on-chain search has become the mainstream model: transaction expressiveness, mempool privacy, low fees, and the lack of an efficient auction mechanism. Among these, low gas fees and high expressiveness are clear goals for general-purpose smart contract chains ), and we need to continue to reinforce these characteristics. Therefore, the solution must focus on the other two points: enabling searchers to read the upcoming on-chain state and expressing their preferences in a way that both protects user rights and minimizes on-chain junk transactions.

Solution Direction

1 Achieving state transparency through programmable privacy

An efficient market needs to provide searchers with real-time access to trading flows while programmatically restricting their use of information. The system must verifiably ensure that searchers can only execute backrun trades and cannot implement frontrun, sandwich attacks, or leak private data. This visibility allows searchers to execute conditional logic off-chain rather than blindly probing on-chain. Once searchers generate potentially profitable trades off-chain, there still needs to be a way to precisely embed them into blocks to capture MEV.

2 Building an explicit bidding MEV auction mechanism

Abandon the "garbage trading auction" model that competes based on Gas consumption, and instead design a transaction ordering right bidding mechanism based on economic incentives. Searchers can directly submit monetary bids for the block position of the target transaction, determining the transaction order through a market-based pricing mechanism. This model transforms the chaotic competition of Gas consumption into an efficient price discovery process:

Searchers do not need to send hundreds of invalid transactions, they only need to pay for the sorting rights that are truly valuable;

Blockchain can capture the true value of MEV through auctions, rather than letting resources waste on meaningless on-chain computations.

Flashbots is attempting to provide visibility to searchers using Trusted Execution Environments (TEEs) while preventing sandwich attacks. TEEs ensure that specific code remains confidential even to the machine operators during execution.

This enables searchers to run verifiably on private transactions in TEE, while being unable to implement sandwich attacks or export any privacy data. We have validated this model on Ethereum L1, where searchers have been conducting post-run transactions through a similar system for several months and are actively adapting it to L2.

Conclusion

For a long time, the discussion of scaling has been limited to the underlying technology throughput. But our research shows that the key breaking point is no longer to expand block capacity, but to make more efficient use of block space [1]. This is because for every unit of block space released, MEV incentivizes spam transactions to consume additional capacity. In other words, most of the revenue from "scaling" is captured by economically rational MEV bots, and real users cannot benefit from it. This problem is pushing up the fees of ordinary users, restricting the effectiveness of capacity expansion, and causing a waste of massive network resources.

The limitation of scalability lies here: increasing block space can enhance throughput, but the improvement in transaction fees is limited, as the increasingly complex on-chain MEV will consume most of the gains. To break through these limitations and unleash the true potential of scalability, we must get rid of the wasteful junk trading market. Through programmable privacy and explicit bidding, we can eliminate the incentives for junk trading, replacing it with an expressive, fair, and efficient MEV market instead of "junk trading auctions."

Adopting MEV auctions is not a luxury choice, but a strategic necessity. The core lies in utilizing TEEs to provide searchers with access to transaction flow while programming restrictions on how they can use it. This design can achieve the ideal outcome: supporting backrunning arbitrage without junk transactions, while preventing sandwich attacks. For the blockchain, this means capturing more revenue in an efficient, junk-free market; for users and developers, lower and stable fees along with genuinely usable capacity will ultimately unlock the full value of scaling.

What will happen when we break through the limitations of garbage trading? When transaction costs are low enough to be almost negligible, what new possibilities will be unlocked? What new applications will emerge? The answer can only be proven through practice.

Thanks to the valuable suggestions from DataAlways, Hasu, Fahim, Danning, dmarz, Nathan, Georgios, Dan, buffalu, Quintus, Tesa, Anika, Brian, Xin, Sam, Eli, Christine, Christoph, Alex, Fred, and many others. Special thanks to Phil, and also to Achal for the design assistance.

Appendix

Heuristic Methods for Identifying Garbage Transactions

To identify garbage trades, we have adopted two heuristic rules:

No token transfer: Does the transaction involve any token transfer? If so, it is not classified as a spam transaction.

Repeated DEX price query: If a transaction initiates at least 4 queries on common DEX price data without executing token transfers, it is classified as a spam transaction.

We believe that at the time of writing this article, these heuristic methods are reliable: any operation involving token transfers typically has actual value for users, while junk trades only transfer tokens when capturing MEV opportunities. Furthermore, the DEX price query rules can effectively identify robots that systematically probe arbitrage opportunities, which is the main form of junk trading we have observed. This definition focuses on wasteful behavior that only queries DEX prices on-chain, excluding productive backrunning behavior.

However, this definition needs further optimization in the future: garbage trading bots can bypass this rule by simply transferring tokens, so the classification criteria for "garbage trading" remains a direction worth further research. In addition, this definition primarily covers the blind back-running arbitrage bots that dominate MEV, and does not include other MEV strategies such as liquidation.

Garbage Trading Identification Methodology

We identify spam transactions by analyzing transaction traces: for each transaction, check all traces of it to determine whether to call a token transfer function or a DEX price function (e.g., slot0[2][3], getReserves(), etc.). If the transaction involves the transfer of tokens, it is excluded; If the tokens are not transferred and 4 or more DEX price queries are initiated, it is classified as a spam transaction.

The choice of 4 times as the threshold was conservative, and experiments have shown that setting the threshold to 3 times has little effect on the overall results. Similarly, we filtered transactions by transfer events on Dune and found that the results were not much different from the track-based approach.

spam-inspect tool

To study spam trading, we developed spam-inspect, a Python tool specifically designed for analyzing Ethereum Rollup activities, aimed at efficiently identifying spam bot behaviors. This tool analyzes by tracking each transaction within the block and implementing the heuristic rules mentioned above.

This tool relies on the trace_block method and is currently only available on OP-Stack chains that support OP-Reth or OP-Erigon.

Dune Query

We built materialized views on Dune by filtering transactions that contain Transfer events and identifying duplicate DEX price calls to locate hash values that meet the criteria for spam transactions. The difference from spam-inspect is that this method relies on transfer events rather than transaction tracking. These spam transaction materialized views are used for subsequent query analysis.

Data Availability (DA) Estimation

Although this article mainly discusses the impact of junk trading pairs on Gas, it will also consume other resources, such as the occupation of L1 data availability by Rollup. To estimate the L1 DA resources wasted by L2 junk trading, we built a custom data pipeline (reusing some modules of op-batcher) and obtained results through two sets of calculations:

Total size after compression of all transactions in the block;

Total size after block compression following the removal of junk transactions.

The difference between the two is the estimated value of L1 DA consumed by junk transactions in a single block.

Footnote

( This indicates that the MEV usage of the chain will expand in tandem with its throughput growth.

) The logic of a specific application chain (app-specific chain) may differ: intentionally limiting transaction expressiveness may be an effective strategy in this scenario.

[1] Explicit auctions solve the systemic inefficiency of resource allocation but introduce a new limitation: the time required to conduct a fair competitive auction. Affected by network latency and the computational load of the auction, this time sets a lower limit on block time, indicating a trade-off between maximizing block space utilization and minimizing block time. A related article will be published soon.