Block-level Access Lists (BALs)
Many thanks to Francesco, Jochem, Vitalik, Ignacio, Gary, Dankrad, Carl, Ansgar, and Tim for feedback and review.
bal_preview
TL;DR: Block builders include access lists...

<div><h1><a href="https://ethresear.ch#p-54283-block-level-access-lists-bals-1" rel="nofollow"></a>Block-level Access Lists (BALs)</h1><blockquote>Many thanks to <a href="https://x.com/fradamt" rel="nofollow">Francesco</a>, <a href="https://x.com/JochemBrouwer96" rel="nofollow">Jochem</a>, <a href="https://x.com/VitalikButerin" rel="nofollow">Vitalik</a>, <a href="https://x.com/ignaciohagopian" rel="nofollow">Ignacio</a>, <a href="https://x.com/gary_rong" rel="nofollow">Gary</a>, <a href="https://x.com/dankrad" rel="nofollow">Dankrad</a>, <a href="https://x.com/CarlBeek" rel="nofollow">Carl</a>, <a href="https://x.com/adietrichs" rel="nofollow">Ansgar</a>, and <a href="https://x.com/TimBeiko" rel="nofollow">Tim</a> for feedback and review.</blockquote><div><a href="https://ethresear.ch/uploads/default/original/3X/a/4/a4bd446262b8310169e42643ac48941f96ff572a.jpeg" title="bal_preview" rel="nofollow"><img src="https://ethresear.ch/uploads/default/optimized/3X/a/4/a4bd446262b8310169e42643ac48941f96ff572a_2_432x375.jpeg" alt="bal_preview"><div>bal_preview1024×888 145 KB</div></a></div>TL;DR: Block builders include access lists and state diffs into blocks for validators to validate faster → scale L1!One of the key topics in the renewed effort to scale Ethereum’s Layer 1—especially the execution layer—is <a href="https://github.com/ethereum/EIPs/blob/master/EIPS/eip-7928.md" rel="nofollow">block-level access lists (BALs) - EIP-7928</a>.BALs are structured lists that the block builder must include in each block, specifying which storage slots individual transactions will access. If these lists are inaccurate or incomplete, the block becomes invalid. As a result, Ethereum’s consensus mechanism can enforce strict compliance, requiring builders to provide correct BALs.Validators benefit significantly from BALs by accelerating block verification. Knowing exactly which accounts and storage slots will be accessed enables validators to apply simple parallelization strategies for disk reads (IO) and execution (EVM). This can lead to faster block validation and open the door to raising gas limits in the future.<blockquote>See Dankrad’s simulations on parallelizing storage reads <a href="https://notes.ethereum.org/@dankrad/SJawZpx1yx" rel="nofollow">here</a>.</blockquote>A critical design goal for BALs is maintaining compact size in both average and worst-case scenarios. Bandwidth is already a constraint for nodes and validators, so it’s vital that BALs don’t add unnecessary load to the network. BALs will be put into the block body, with a hash of the BAL stored in the header.<blockquote>Currently, Ethereum clients rely on their own optimistic parallelization strategies. These generally perform well on average blocks, but struggle with worst-case scenarios, leading to significant performance differences.</blockquote><h1><a href="https://ethresear.ch#p-54283-the-design-space-for-bals-2" rel="nofollow"></a>The design space for BALs</h1><ul><li>What to include?<ul><li>In addition to storage locations <code>(address, storage key)</code>, we can also include:<ul><li>storage values</li><li>balances</li><li>balance diffs</li></ul></li><li>For storage values, we can distinguish between:<ul><li>pre-block vs. pre-transaction execution values</li><li>pre- vs. post-transaction execution values</li></ul></li><li>Are we aiming to be hardware-agnostic, or do we want to optimize for certain commonly used hardware specs?</li></ul></li></ul>In the following, we’ll focus on three main variants of BALs: access, execution and state.<h1><a href="https://ethresear.ch#p-54283-access-bals-3" rel="nofollow"></a>Access BALs</h1>Access BALs map transactions to <code>(address, storage key)</code> tuples.<ul><li>Small in size.</li><li>Enable parallel disk reads but are less effective for parallel transaction validation due to potential dependencies.</li><li>Execution time is <code>parallel IO + serial EVM</code>.</li></ul>For efficiency, a lightweight BAL structure could look like this using SSZ:<pre><div><button></button><button></button></div><code># Type aliasesAddress = ByteVector(20)StorageKey = ByteVector(32)TxIndex = uint16# Constants; chosen to support a 630m block gas limitMAX_TXS = 30_000MAX_SLOTS = 300_000MAX_ACCOUNTS = 300_000# Containersclass SlotAccess(Container):slot: StorageKeytx_indices: List[TxIndex, MAX_TXS] # Transactions (by index) that access this slot (read or write)class AccountAccess(Container):address: Addressaccesses: List[SlotAccess, MAX_SLOTS]# Top-level block fieldsBlockAccessWrites = List[AccountAccess, MAX_ACCOUNTS]BlockAccessReads = List[AccountAccess, MAX_ACCOUNTS]</code></pre>The outer list is a deduplicated list of addresses accessed during the block.<ul><li>For each address, there’s a list of storage keys accessed.</li><li>For each storage key:<ul><li><code>List[TxIndex]</code>: Ordered transaction indices that accessed this key.</li></ul></li></ul>For example, the BAL for block number <code>21_935_797</code> would look like this:<pre><div><button></button><button></button></div><code>[('0xc02aaa39b223fe8d0a0e5c4f27ead9083c756cc2',[('0xa63b...c2', [0]),('0x8b3e...a7', [0]),('0xfb19...a8',[1, 2, 3, 4, 84, 85, 91]),# ... additional entries]),# ... additional entries]</code></pre><h1><a href="https://ethresear.ch#p-54283-execution-bals-4" rel="nofollow"></a>Execution BALs</h1>Execution BALs map transactions to <code>(address, storage key, value)</code> tuples and include balance diffs.<ul><li>Slightly larger in size due to value inclusion.</li><li>Facilitate parallel disk reads and parallel execution.</li><li>Execution time is <code>parallel IO + parallel EVM</code>.</li></ul>An efficient structure using SSZ:<pre><div><button></button><button></button></div><code># Type aliasesAddress = ByteVector(20)StorageKey = ByteVector(32)StorageValue = ByteVector(32)TxIndex = uint16Nonce = uint64# Constants; chosen to support a 630m block gas limitMAX_TXS = 30_000MAX_SLOTS = 300_000MAX_ACCOUNTS = 300_000MAX_CODE_SIZE = 24576 # Maximum contract bytecode size in bytes# SSZ containersclass PerTxAccess(Container):tx_index: TxIndexvalue_after: StorageValue # value in state after the last access within the transactionclass SlotAccess(Container):slot: StorageKeyaccesses: List[PerTxAccess, MAX_TXS] # empty for readsclass AccountAccess(Container):address: Addressaccesses: List[SlotAccess, MAX_SLOTS]code: Union[ByteVector(MAX_CODE_SIZE), None] # Optional field for contract bytecodeBlockAccessList = List[AccountAccess, MAX_ACCOUNTS]# Pre-block nonce structuresclass AccountNonce(Container):address: Address # account addressnonce_before: Nonce # nonce value before block executionNonceDiffs = List[AccountNonce, MAX_TXS]</code></pre>The structure is the same as in the access version, with <code>StorageValue</code> added to represent the value after the last access by each transaction.<ul><li>Exclude <code>SlotAccess.accesses</code> for reads: empty <code>SlotAccess.accesses</code> indicates a read.<ul><li>This means that only write operations consist of <code>(StorageKey, List[TxIndex], StorageValue)</code> tuples, significantly reducing the object’s size.</li></ul></li><li>Instead of post-execution values, we could include pre-execution values for reads and writes for each transaction. Thereby, EVM execution would not have to wait for disk reads. This is a completely separate design, coming with its own trade-offs, reducing execution time to <code>max(parallel IO, parallel EVM)</code>.</li></ul><blockquote>For syncing (c.f. healing phase), having state diffs (thus, the post-tx values) for writes is required for catching up with the chain while updating state. Instead of receiving new state values directly with their proofs, we can heal state using the state diffs inside blocks and verify the correctness of the process by comparing the final derived state root to the head block’s state root received from a light node (h/t dankrad).</blockquote>For contract deployments, the <code>code</code> must contain the runtime bytecode of the newly deployed contract.The <code>NonceDiffs</code> structure MUST record the pre-transaction nonce values for all <code>CREATE</code> and <code>CREATE2</code> deployer accounts and the deployed contracts in the block.An example BAL for block <code>21_935_797</code> might look like this:<pre><div><button></button><button></button></div><code>[('0xc02aaa39b223fe8d0a0e5c4f27ead9083c756cc2',[('0xa63b...c2',[0],'0x...'),('0x8b3e...a7',[0],'0x...'),('0xfb19...a8',[1, 2, 3, 4, 84, 85, 91],'0x...'),...])]</code></pre><hr>Balance diffs are needed to correctly handle execution that depends on an account’s balance. These diffs include every address touched by a transaction involving value transfers, along with the balance deltas, transaction senders, recipients, and the block’s coinbase.<pre><div><button></button><button></button></div><code># Type aliasesAddress = ByteVector(20)TxIndex = uint64BalanceDelta = ByteVector(12) # signed, two's complement encoding# ConstantsMAX_TXS = 30_000MAX_ACCOUNTS = 70_000 # 630m / 9300 (cost call to non-empty acc with value)# Containersclass BalanceChange(Container):tx_index: TxIndexdelta: BalanceDelta # signed integer, encoded as 12-byte vectorclass AccountBalanceDiff(Container):address: Addresschanges: List[BalanceChange, MAX_TXS]BalanceDiffs = List[AccountBalanceDiff, MAX_ACCOUNTS]</code></pre><ul><li>Deduplicated per address.</li><li>Each tuple lists the exact balance change for every relevant transaction.</li></ul>Example:<pre><div><button></button><button></button></div><code>[('0xdead...beef',[(0, -1000000000000000000), # tx 0: sent 1 ETH(2, +500000000000000000) # tx 2: received 0.5 ETH]),# ... additional entries]</code></pre><h1><a href="https://ethresear.ch#p-54283-state-bals-5" rel="nofollow"></a>State BALs</h1>This structure fully decouples execution from state, allowing validators to bypass any disk or trie lookups during execution, relying solely on the data provided in the block. The <code>pre_accesses</code> list provides the initial values of all accessed slots before the block starts, while <code>tx_accesses</code> traces the per-transaction access patterns and post-access values, enabling fine-grained parallel execution and verification.<ul><li>Larger in size</li><li>Execution time is <code>max(parallel IO, parallel EVM)</code>.</li></ul>An efficient SSZ object could look like the following:<pre><div><button></button><button></button></div><code># Type aliasesAddress = ByteVector(20)StorageKey = ByteVector(32)StorageValue = ByteVector(32)TxIndex = uint16# ConstantsMAX_TXS = 30_000MAX_SLOTS = 300_000MAX_ACCOUNTS = 300_000# Sub-containersclass PerTxAccess(Container):tx_index: TxIndexvalue_after: StorageValueclass SlotAccess(Container):slot: StorageKeyaccesses: List[PerTxAccess, MAX_TXS]class AccountAccess(Container):address: Addressaccesses: List[SlotAccess, MAX_SLOTS]class SlotPreValue(Container):slot: StorageKeyvalue_before: StorageValueclass AccountPreAccess(Container):address: Addressslots: List[SlotPreValue, MAX_SLOTS]# Unified top-level containerclass BlockAccessList(Container):pre_accesses: List[AccountPreAccess, MAX_ACCOUNTS]tx_accesses: List[AccountAccess, MAX_ACCOUNTS]</code></pre><blockquote>The balance and nonce diff remains the same. …</blockquote><h3><a href="https://ethresear.ch#p-54283-what-about-excluding-the-initial-read-values-6" rel="nofollow"></a>What about excluding the initial read values?</h3>Another variant of State BAL excludes read values, and only includes pre- and post-values for writes. In this model, <code>pre_accesses</code> and <code>tx_accesses</code> only contain storage slots that were written to, along with their corresponding <code>value_before</code> (from state) and <code>value_after</code> (from the transaction result).This reduces the size while still enabling full state reconstruction, as write slots define the only persistent changes. Read accesses are implicitly assumed to be resolved via traditional state lookups or cached on the client.<h1><a href="https://ethresear.ch#p-54283-worst-case-sizes-7" rel="nofollow"></a>Worst-Case Sizes</h1><h2><a href="https://ethresear.ch#p-54283-access-bal-8" rel="nofollow"></a>Access BAL</h2>The worst-case transaction consumes the entire block gas limit (36 million, as of April 2025) to access as many storage slots inside different contracts as possible by including them in the EIP-2930 access list.Thus, <code>(36_000_000 - 21_000) // 1900</code> gives us the max number of addresses (20 bytes) + storage keys (32 bytes) reads we can do, resulting in <code>18_947</code> storage reads and approximately 0.93 MiB.<blockquote>This is a pessimistic measure. It’s practically infeasible to use the block’s gas exclusively for SLOADs. With a custom contract (see <a href="https://github.com/nerolation/ethereum-stresser-contracts/blob/main/contracts/sload_stresser.sol" rel="nofollow">here</a>), I was able to trigger 16,646 SLOADs, not more. See <a href="https://dashboard.tenderly.co/tx/0x85bbcc2690fbf0ae9af5af1b3d4c006fd9804644695f095cba9e00ff09eaae8b" rel="nofollow">this example transaction on sepolia</a>.</blockquote>This is less than the current (and post-Prectra) worst-case block size achievable through calldata.<blockquote>Average BAL size sampled over 1,000 blocks between 21,605,993 and 2,223,0023 was around 57 KiB SSZ-encoded. On average, blocks in that time frame contained around 1,181 storage keys and 202 contracts per block.</blockquote><h2><a href="https://ethresear.ch#p-54283-execution-bals-9" rel="nofollow"></a>Execution BALs</h2>Including a 32-byte value per write entry doesn’t increase the worst-case BAL size. For reading <code>18_947</code> storage loads, it remains at 0.93 MiB.Worst-case balance diffs occur if a single transaction sends minimal value (1 wei) to multiple addresses:<ul><li>With the <code>21,000</code> base cost and <code>9,300</code> gas for calls, we get a maximum of 3,868 called addresses in one transaction (<code>(36_000_000-21_000)/9_300</code>). Including the <code>tx.from</code> and <code>tx.to</code> of that transaction + the block’s coinbase, we get <code>3,871</code> addresses. With 20-byte addresses and 12-byte balance deltas, we get a balance diff size of 0.12 MiB (12 bytes are enough to represent the total ETH supply).</li><li>Alternatively, using multiple transactions each sending 1 wei to a different account, we can theoretically pack 1,188 transactions into one block. With 3 addresses (<code>callee</code>, <code>tx.from</code> and <code>tx.to</code>) in the balance diff, and 12-byte deltas, we get 0.12 MiB in size.</li></ul><blockquote>A balance diff size sampled over 1,000 blocks (2,160,5993–2,223,0023) would have contained around 182 transactions and 250 addresses with balance deltas on average. This results in an average of 9.6 KiB, SSZ-encoded.</blockquote><h2><a href="https://ethresear.ch#p-54283-state-bals-10" rel="nofollow"></a>State BALs</h2>Including another 32 bytes for reads and writes increased the worst-case BAL size to around 1.51 MiB.<h1><a href="https://ethresear.ch#p-54283-comparison-of-different-designs-11" rel="nofollow"></a>Comparison of different designs</h1><div><a href="https://ethresear.ch/uploads/default/original/3X/a/f/af66c856b222706bd7ab1dcbbf34b088c78577b3.png" title="Screenshot from 2025-04-14 20-36-18" rel="nofollow"><img src="https://ethresear.ch/uploads/default/optimized/3X/a/f/af66c856b222706bd7ab1dcbbf34b088c78577b3_2_651x500.png" alt="Screenshot from 2025-04-14 20-36-18"><div>Screenshot from 2025-04-14 20-36-18904×694 102 KB</div></a></div>The current design specified in <a href="https://github.com/ethereum/EIPs/blob/master/EIPS/eip-7928.md" rel="nofollow">EIP-7928</a> adopts the Execution BAL model with post-tx values. This variant offers a compelling trade-off: it enables both I/O and EVM parallelism and includes sufficient state diffs for syncing, without the additional bandwidth costs associated with pre-transaction state snapshots.<h2><a href="https://ethresear.ch#p-54283-references-12" rel="nofollow"></a>References</h2><ul><li><a href="https://docs.google.com/document/d/1FoW_iuIPoYAjxy-AUGr_vRbIK_Zon3XDAb6heS_GWL0/edit?usp=sharing" rel="nofollow">ESP Academic Grant 2023: EthStorage/Quarkchain Earlier BAL Submission</a></li><li><a href="https://hackmd.io/X4Z4h-EQRPSiQF38rpN9aQ?view" rel="nofollow">Initial tests and simulations by EthStorage/Quarkchain</a></li><li><a href="https://www.scs.stanford.edu/24sp-cs244b/projects/Concerto_Transaction_Parallel_EVM.pdf" rel="nofollow">https://www.scs.stanford.edu/24sp-cs244b/projects/Concerto_Transaction_Parallel_EVM.pdf</a></li><li><a href="https://www.microsoft.com/en-us/research/wp-content/uploads/2021/09/3477132.3483564.pdf" rel="nofollow">https://www.microsoft.com/en-us/research/wp-content/uploads/2021/09/3477132.3483564.pdf</a></li><li><a href="https://writings.flashbots.net/speeding-up-evm-part-1" rel="nofollow">Speeding up the EVM (part 1) | Flashbots Writings</a></li></ul><div> </div></div>

Block-level Access Lists (BALs)

블록 수준 접근 목록(밸런서(BAL))
프란체스코, 요헴, 비탈릭, 이그나시오, 게리, 단크라드, 칼, 안스가르, 팀에게 피드백과 검토에 대해 깊이 감사드립니다.
bal_preview
요약: 블록 빌더들은 접근 목록을 포함합니다...

<div><h1><a href="https://ethresear.ch#p-54283-block-level-access-lists-bals-1" rel="nofollow"></a>블록 수준 접근 목록 (BALs)</h1><blockquote>프란체스코, 요헴, 비탈릭, 이그나시오, 게리, 단크라드, 칼, 앙스가르, 팀에게 피드백과 검토에 대한 깊은 감사를 드립니다.</blockquote><div><a href="https://ethresear.ch/uploads/default/original/3X/a/4/a4bd446262b8310169e42643ac48941f96ff572a.jpeg" title="bal_preview" rel="nofollow"><img src="https://ethresear.ch/uploads/default/optimized/3X/a/4/a4bd446262b8310169e42643ac48941f96ff572a_2_432x375.jpeg" alt="bal_preview"><div>bal_preview1024×888 145 KB</div></a></div>요약: 블록 빌더가 접근 목록과 상태 차이를 블록에 포함시켜 검증자가 더 빠르게 검증할 수 있도록 함 → L1 확장!이더리움의 레이어 1을 확장하기 위한 노력, 특히 실행 계층에서 핵심 주제 중 하나는 <a href="https://github.com/ethereum/EIPs/blob/master/EIPS/eip-7928.md" rel="nofollow">블록 수준 접근 목록(BALs) - EIP-7928</a>입니다.

(이하 생략, 전체 번역 가능합니다. 원하시는 부분의 전체 번역을 원하시면 알려주세요.)<pre><div><button></button><button></button></div><code># 유형 별칭주소 = 바이트 벡터(20)거래 인덱스 = uint64잔액 변동 = 바이트 벡터(12) # 부호 있는, 2의 보수 인코딩# 상수최대_거래 = 30_000최대_계정 = 70_000 # 630m / 9300 (값이 있는 비어있지 않은 계정에 대한 호출 비용)# 컨테이너클래스 잔액변경(컨테이너):거래_인덱스: 거래 인덱스변동: 잔액 변동 # 부호 있는 정수, 12바이트 벡터로 인코딩됨클래스 계정잔액차이(컨테이너):주소: 주소변경: 목록[잔액변경, 최대_거래]잔액차이 = 목록[계정잔액차이, 최대_계정]</code></pre><ul><li>주소별로 중복 제거됨.</li><li>각 튜플은 모든 관련 거래의 정확한 잔액 변경을 나열함.</li></ul>예시:<pre><div><button></button><button></button></div><code>[('0xdead...beef',[(0, -1000000000000000000), # 거래 0: 1 이더리움(ETH) 전송(2, +500000000000000000) # 거래 2: 0.5 이더리움(ETH) 수신]),# ... 추가 항목]</code></pre><h1><a href="https://ethresear.ch#p-54283-state-bals-5" rel="nofollow"></a>상태 잔액</h1>이 구조는 실행을 상태와 완전히 분리하여 검증자가 블록에서 제공된 데이터에만 의존하면서 디스크나 트라이 조회를 우회할 수 있게 합니다. <code>pre_accesses</code> 목록은 블록 시작 전에 접근된 모든 슬롯의 초기 값을 제공하고, <code>tx_accesses</code>는 거래별 접근 패턴과 접근 후 값을 추적하여 세분화된 병렬 실행 및 검증을 가능하게 합니다.<ul><li>크기가 더 큼</li><li>실행 시간은 <code>max(병렬 IO, 병렬 이더리움 가상 머신(EVM))</code>입니다.</li></ul>효율적인 SSZ 객체는 다음과 같을 수 있습니다:

(이하 생략, 동일한 방식으로 번역됨)

블록 수준 액세스 목록(BAL)

암호화폐 거래자들이 소셜 미디어에서 시장 바닥을 확인할 만큼 충분한 공포감을 아직 보이지 않았다고 한 암호화폐 분석가는 말하며, 비트코인 가격이 여전히 7만 5천 달러 부근까지 하락할 수 있다고 시사했습니다.
"매우 유혹적으로 보이네요..."

암호화폐 시장은 아직 바닥을 쳤다고 단정짓기에는 ‘충분히 겁먹지 않았다’: 산티먼트

ME 뉴스에 따르면, 12월 21일(UTC+8) 이더 창시자 비탈릭 부테린은 파캐스터(Farcaster)에 글을 올려 예측 시장이 정서 주제에 대한 비이성적인 관점 에 대한 좋은 해결책이 될 수 있다고 주장했습니다. 그는 두 개의 스크린샷을 예시로 들었는데, 일론 머스크가 이전에 영국 내전이 "불가피하다"고 주장했던 것을 보여주면서 예측 시장 폴리마켓(Polymarket)에서는 "2024년에 영국 내전이 일어날 것인가?"라는 질문에 대해 3%의 확률만 제시했다는 것입니다. (비탈릭은 일부 베팅 참여자들이 확률을 부풀렸기 때문에 3%도 너무 높다고 생각합니다.) 비탈릭은...

비탈릭: 예측 시장은 정서 주제에 대한 과격한 관점 분산시켜 소셜 미디어의 "해결책"이 될 수 있습니다.

한 암호화폐 사용자가 대규모 온체인 공격으로 인한 주소 오염 사기에 속아 5천만 달러 상당의 테더 USDT(USDT) 잃었습니다.
웹3 보안 회사인 웹3 안티바이러스가 발견한 이 도용 사건은 사용자가 메시지를 보낸 후에 발생했습니다...