Move committee index outside Attestation

Abstract

Move the committee index field outside of the signed Attestation message to allow aggregation of equal consensus votes.

Motivation

This proposal aims to make Casper FFG clients more efficient by reducing the average number of pairings needed to verify consensus rules. While all types of clients can benefit from this EIP, ZK circuits proving Casper FFG consensus are likely to have the most impact.

On a beacon chain network with at least 262144 active indexes it's necessary to verify a minimum of ceil(32*64 * 2/3) = 1366 attestations to reach a 2/3 threshold. Participants cast two votes at once: LMD GHOST vote and Casper-FFG vote. However, the Attestation message contains three elements:

1. LMD GHOST vote (beacon_block_root, slot). Note: includes slot in the event (block, slot) voting is adopted.
2. FFG vote (source, target)
3. Committee index (index)

Signing over the 3rd item causes tuples of equal votes to produce different signing roots. If the committee index is moved outside of the Attestation message the minimum number of attestations to verify to reach a 2/3 threshold is reduced to ceil(32 * 2/3) = 22 (a factor of 62).

On-chain attestations can now be packed more space-efficiently into beacon blocks. This proposal allows to include up to 8 slots worth of votes into a block, compared to 2 today. In order words, a chain with only 1/8 online proposers can still potentially include all votes on-chain.

Specification

Execution layer

This requires no changes to the Execution Layer.

Consensus layer

• Set index field from AttestationData to fixed value of zero
• Move committee indexing data to the outter Attestation container with committee_bits
• Increase the capacity of aggregation_bits to all committees in a slot

The full specification of the proposed change can be found in /specs/electra/beacon-chain.md.

Rationale

Deprecation strategy

The index field in AttestationData can be deprecated by:

1. Removing the field
2. Preserving the field and setting it to be zero
3. Changing the field type to Optional (from EIP-7495 StableContainer)

This EIP chooses the second option to not complicate the inclusion of AttesterSlashing objects. While the Attestation container changes, AttesterSlashing includes indexed attestations without committee data.

MAX_ATTESTATIONS value

The maximum size of an attestation increases, with a bitfield 64 times larger on networks with maxed committees. MAX_ATTESTATIONS value is reduced to limit the beacon block size while still increase the total capacity of votes. A value of 8 increase the vote capacity by 4 while having the same attestation space size with a network of 1.2M active indices.

MAX_ATTESTER_SLASHINGS value

On-chain AttesterSlashing includes a list of indices of all participants. With this EIP the worst-case size increases by 64 times, resulting in an uncompressed size of 488 KB per AttesterSlashing in a network of 1M validators. Snappy compression reduces it to 320 KB, which is still significant. To bound the maximum size of block this proposal reduces MAX_ATTESTER_SLASHINGS from 2 to 1, the minimum value.

Backwards Compatibility

This EIP introduces backward incompatible changes to the block validation rule set on the consensus layer and must be accompanied by a hard fork.

Security Considerations

First block after fork

Because the on chain Attestation container changes, attestations from the prior fork can't be included into post-electra blocks. Therefore the first block after the fork may have zero attestations. LMD votes can still be applied to fork-choice via on_attestation handler, so there will be only a 1/32 loss of FFG votes. Attesters assigned to the last slot of the fork will incur one epoch worth of offline penalties. One possible mitigation is to change the electra block body type to allow including attestations from both forks. However, the mitigation increases complexity for little gain so this proposal choses to not address the issue.

Mutation over gossip

Moving the index field outside of the signed message allows malicious mutation only on the p2p gossip topic beacon_attestation_${subnet_id}. Everywhere else, the Attestation message is wrapped with an outer signature that prevents mutation.

Gossip verification rules for the beacon_attestation_${subnet_id} topic include:

• [IGNORE] There has been no other valid attestation seen on an attestation subnet that has an identical attestation.data.target.epoch and participating validator index.
• [REJECT] The signature of attestation is valid.

For an unaggregated attestation, the tuple (slot, index, aggregation_bits) uniquely identifies a single public key. Thus there is a single correct value for the field index. If an attacker mutates the index field the signature will fail to verify and the message will be dropped. This is the same outcome of mutating the aggregation bits, which is possible today. If implementations verify the attestation signature before registering it in a 'first-seen' cache, there's no risk of cache pollution.

Copyright

Copyright and related rights waived via CC0.