Add EIP: eth/vhash - Blob-Aware Mempool

EIPS/eip-8094.md

@@ -0,0 +1,179 @@
+---
+eip: 8094
+title: eth/vhash - Blob-Aware Mempool
+description: Make mempool messaging vhash aware
+author: Csaba Kiraly (@cskiraly)
+discussions-to: https://ethereum-magicians.org/t/eip-8094-eth-vhash-blob-aware-mempool/26834
+status: Draft
+type: Standards Track
+category: Networking
+created: 2025-11-29
+requires: 7642
+---
+
+## Abstract
+
+This EIP eliminates the need to redistribute blob content in the mempool if
+only the metadata (fees) of a transaction are updated, making RBF
+(replace-by-fee) more efficient and cheaper for the network. It achieves this
+modifying the devp2p ‘eth’ protocol to address blobs in type3 transaction
+sidecars by content (vhash).
+
+## Motivation
+
+In the current version of devp2p eth/69, when a transaction is replaced, it
+must be redistributed in the mempool like any new transaction. Even if the
+actual content is largely the same, protocol participants have no means to
+figure this out before getting the full content, making a replacement use the
+same amount of network resources as a new transaction would.
+
+What is especially problematic is that RBF is used most in periods of fee
+volatility, and a network overload is the typical case of such a situation.
+Thus, when there is already high demand, we make the situation worse by adding
+extra traffic redistributing blob content that was already distributed.
+
+## Specification
+
+### Transactions (0x02) changes
+
+Type 3 transaction should be sent without sidecar
+
+### PooledTransactions (0x0a) changes
+
+Type 3 transaction should be sent without sidecar
+
+### GetPooledBlobs (msg code TBA)
+
+[request-id: P, [vhash₁: B_32, vhash₂: B_32, ...]]
+
+This message requests blobs from the recipient's transaction pool by vhash.
+
+### PooledBlobs (msg code TBA)
+
+[request-id: P, [blob₁, blob₂...]]
+
+This is the response to GetPooledBlobs, returning the requested blobs. The
+items in the list are blobs in the format described in the main Ethereum
+specification.
+
+> Note: optionally, we might prefix the blob format with the blob version number
+>
+> Note: optionally, we might decide to improve the blob format allowing nodes
+to reconstruct RS encoding instead of using extra bandwidth, by sending the
+following fields per blob:
+>
+> - blob version
+> - blob content, **excluding** the erasure coding extension in case of version 1
+> - blob commitment
+> - blob proof(s), **including** cell proofs of the erasure coded piece in case
+of version 1
+>
+> It is important to include all cell proofs to keep reconstruction
+CPU-efficient.
+
+The blobs must be in the same order as in the request, but it is OK to skip
+blobs which are not available. Since the recipient have to check that
+transmitted blob hashes correspond to the requested vhashes anyway, we can
+avoid sending the list of vhashes as part of this message.
+
+> Note: Optionally, we could extend this message with a bitmap of sent/unsent
+blobs from the request, or with the list of vhashes sent. This information is
+redundant, but it can simplify processing on the receiver side.
+
+### Other spec changes
+
+EIP-4844 introduced the following:
+
+    Nodes MUST NOT automatically broadcast blob transactions to their peers.
+    Instead, those transactions are only announced using
+    NewPooledTransactionHashes messages, and can then be manually requested
+    via GetPooledTransactions.
+
+The above should be changed as follows:
+
+    Nodes MUST send (broadcast or send in NewPooledTransactionHashes) blob
+    transactions **without** sidecars to their peers. Peers can then request
+    blob content using `GetPooledBlobs` messages. Nodes MUST NOT forward blob
+    transactions before receiving and validating all blobs".
+
+## Rationale
+
+A typical blob transaction RBF changes the fees only, while the sidecar (blob
+content) remains the same. If a node that has the previous version would know
+this, it could avoid pulling the sidecar, largely reducing bandwidth
+consumption. However, this is not possible with the current messaging. To make
+this happen, we have to expose blob (or at least sidecar) identifiers in
+mempool messaging.
+
+### Should we use blob identifiers or a sidecar identifier?
+
+We can either use vhashes, or a sidecar level hash. The latter has the slight
+advantage of being a single element, thus simplifying message format, but it
+has several disadvantages:
+
+- It would be a new identifier, while the blob level vhash is already well
+established (just not in devp2p)
+- It would not allow restructuring the message, sending e.g. less blobs under a
+fee surge
+
+Thus, we use vhashes.
+
+### Implementation options
+
+There are several options to bring vhashes to devp2p messaging:
+
+#### Option 1
+
+- Extend announcements with vhashes
+- Allow nodes to request transaction with/without sidecar content, or even
+selecting which parts are needed (bitmap or vhash list)
+
+#### Option 2
+
+- Extend announcements with nonce (see EIP-8077)
+- If hash differs from what we have, request with/without sidecar content based
+on whether we already have the sidecars for the same nonce, assuming this is a
+simple replacement
+- Request again with sidecar if vhashes differ
+
+#### Option 3 (selected)
+
+- Push (or announce and then pull) type 3 transaction **without sidecar**
+- Allow to **request sidecar separately** (new message type)
+
+At first it might seem that Option 3 is slowing down distribution, adding one
+more RTT latency per hop. However, since most type 3 transactions are small
+without a sidecar, we could change the protocol behaviour to allow pushing
+these transactions without the sidecar, leaving it to the receiver of the push
+to ask for the blobs if needed. Forwarding of type3 transactions without
+sidecar should be prohibited until sidecars are fetched and the content can be
+verified.
+
+After considering the above options, we chose to propose Option 3, introducing
+a new message type.
+
+### Relation to other EIPs in draft state
+
+- EIP-8077 (announce source and nonce): the changes can be simply combined.
+- EIP-8070 (Sparse blobpool): both EIPs change how blob transactions are
+propagated over the network. The goal of the two EIPs are different. EIP-8070
+is about a proportional bandwidth reduction in the normal case without dealing
+with specifics of RBF. This EIP is about enabling RBF without using extra
+bandwidth. The two EIPs can be combined, but the combination depends on the
+order of introduction, hence we leave this for later.
+
+## Backwards Compatibility
+
+This EIP changes the eth protocol and requires rolling out a new version.
+Supporting multiple versions of a wire protocol is routine practice. Rolling
+out this new version does not break older clients, since they can keep using
+the previous protocol version.
+
+This EIP does not change consensus rules of the EVM and does not require a hard
+fork.
+
+## Security Considerations
+
+## Copyright
+
+Copyright and related rights waived via CC0.