Enhancing Ethereum PoA Clique Network with DAG-based BFT Consensus

ICBC 2024 Presentation Slides

1. 1
Pohang University of Science And Technology
YongraeJo
Enhancing Ethereum PoAClique Network with DAG-
based BFT Consensus
YongraeJo and ChanikPark

2. 2
lConsensus for permissioned blockchain
─Only a few messages for block agreement
─Efficient than proof-of-work (PoW)
Proof of Authority (PoA)
PoA
Consensus
Authorities
12
3
45
6
Block
Proposal

3. 3
lConsensus for permissioned blockchain
─Only a few messages for block agreement
─Efficient than proof-of-work (PoW)
Proof of Authority (PoA)
PoA
Consensus
Authorities
12
3
45
6
Block
Proposal
Clique
(our focus)

4. 4
PoAClique: An Illustration
AuthoritiesRound 1Round 2Round 3Round 4
Block proposal
2
1
3
4
5
by leader
by non-leader
lRound-based
─(rotating) leader & round timeout Δ
lMultiple block proposals in a round
─by a leader
─by non-leaders* with random delays
*#non-leaders = up to !
"−2

5. 5
PoAClique: An Illustration
AuthoritiesRound 1Round 2Round 3Round 4
2
1
3
4
5
lRound-based
─(rotating) leader & round timeout Δ
lMultiple block proposals in a round
─by a leader
─by non-leaders* with random delays
lSingle block acceptance in a round
─Normal: a leader block
*#non-leaders = up to !
"−2
Block proposal
by leader
by non-leader
Accepted

6. 6
PoAClique: An Illustration
AuthoritiesRound 1Round 2Round 3Round 4
2
1
3
4
5
lRound-based
─(rotating) leader & round timeout Δ
lMultiple block proposals in a round
─by a leader
─by non-leaders* with random delays
lSingle block acceptance in a round
─Normal: a leader block
Block proposal
by leader
by non-leader
Accepted
Delayed
Block
Forks
happen
*#non-leaders = up to !
"−2

7. 7
PoAClique: An Illustration
AuthoritiesRound 1Round 2
2
Round 3Round 4
12
1
2
1
3
4
5
lRound-based
─(rotating) leader & round timeout Δ
lMultiple block proposals in a round
─by a leader
─by non-leaders* with random delays
lSingle block acceptance in a round
─Normal: a leader block
─Forks: a highest scored chain
Block proposal
by leader
by non-leader
Accepted
Scoring mechanism
-leader’s: 2
-non-leaders’: 1
Forks
happen
Delayed
Block
*#non-leaders = up to !
"−2

8. 8
PoAClique: An Illustration
AuthoritiesRound 1Round 2
2
Round 3Round 4
12
12
12
1
2
1
3
4
5
lRound-based
─(rotating) leader & round timeout Δ
lMultiple block proposals in a round
─by a leader
─by non-leaders* with random delays
lSingle block acceptance in a round
─Normal: a leader block
─Forks: a highest scored chain
Block proposal
by leader
by non-leader
Accepted
Delayed
Block
*#non-leaders = up to !
"−2

9. 9
PoAClique: An Illustration
AuthoritiesRound 1Round 2
2
Round 3Round 4
2
12
2
2
1
3
4
5
lRound-based
─(rotating) leader & round timeout Δ
lMultiple block proposals in a round
─by a leader
─by non-leaders* with random delays
lSingle block acceptance in a round
─Normal: aleader block
─Forks: a highest scored chain
Block proposal
by leader
by non-leader
Accepted
-Not proposed
-Invalid block
-Beyond Δ
*#non-leaders = up to !
"−2

10. 10*#non-leaders = up to !
"−2
PoAClique: Issues
AuthoritiesRound 1Round 2
2
Round 3Round 4
2
12
2
2
1
3
4
5
lRound-based
─(rotating) leader & round timeout Δ
lMultiple block proposals in a round
─by a leader
─by non-leaders* with random delays
lSingle block acceptance in a round
─Normal: a leader block
─Forks: a highest scored chain
Block proposal
by leader
by non-leader
Accepted
(network fault)
-Not proposed
-Invalid block
-Beyond Δ
Multiple block proposals, but a single block acceptance

11. 11*#non-leaders = up to !
"−2
PoAClique: Issues
AuthoritiesRound 1Round 2
2
Round 3Round 4
2
12
2
2
1
3
4
5
lRound-based
─(rotating) leader & round timeout Δ
lMultiple block proposals in a round
─by a leader
─by non-leaders* with random delays
lSingle block acceptance in a round
─Normal: a leader block
─Forks: a highest scored chain
Block proposal
by leader
by non-leader
Accepted
(network fault)
-Not proposed
-Invalid block
-Beyond Δ
Limited throughput & Wasted resources
Multiple block proposals, but a single block acceptance

12. 12
Our Idea
AuthoritiesRound 1Round 2Round 3Round 4
2
1
3
4
5
lRound-based
─leader-less& round timeout Δ
lMultiple block proposals in a round
─By all authorities
lMultiple block acceptance in a round
─Normal: (consistent) all proposedblocks
─Forks: (consistent) total order
Block proposal
by leader
by non-leader
Accepted

13. 13
Our Idea
AuthoritiesRound 1Round 2Round 3Round 4
2
1
3
4
5
lRound-based
─leader-less& round timeout Δ
lMultiple block proposals in a round
─By all authorities
lMultiple block acceptance in a round
─Normal: (consistent) all proposedblocks
─Forks: (consistent) total order
Block proposal
by leader
by non-leader
Accepted Challenges

14. 14
Challenge (1) Consistent view across nodes
Authority 1’s view
AuthoritiesRound 1Round 2Round 3Round 4
2
1
3
4
5
X
AuthoritiesRound 1Round 2Round 3Round 4
2
1
3
4
5
Y
Authority 2’s view
Fork
attack
Delayed/
Dropped

15. 15
Challenge (2) Consistent total ordering
Authority 1’s view
AuthoritiesRound 1Round 2Round 3Round 4
2
1
3
4
5
AuthoritiesRound 1Round 2Round 3Round 4
2
1
3
4
5
Authority 2’s view
Authority 1’s
total order?Authority 2’s
total order

16. 16
DAG-based BFT Consensus
─Consistent blocks across nodes
─Reliable block distribution
─Parallel block proposals & Acceptance
DAG MempoolConsensus
Round 1Round 2Round 3Round 4
2
1
3
4
5
Round 1Round 2Round 3Round 4
─Consistent total order
─Local & deterministic
─(shared) randomness

17. 17
Contributions
DAG Mempool
Multiple Blocks AcceptanceDAG-based BFT Consensus
2
1
3
4
5
Single Block Acceptance
Consensus
DaPoAPoAClique
2
1
3
4
5
achieves 2.47x higher throughput, 5.76x lower latency than Clique

18. 18
lA consortium blockchain
─Known Identities
─Members operates DaPoAnode
─Clients submit Txs to a trusted node
─fByzantine members out of 3f+1members
lPartially asynchronous network
─unknown time bound ∆
lCrypto cannot besubverted
Assumptions

19. 19
DaPoA: Operations
Tx
Req.TxPool-Block
-round r
…
DaPoA
Nodes

20. 20
DaPoA: Operations
Tx
Req.TxPoolround r-1round r
…
…
…
…
-Block
-round r
…
…
DAG Mempool
(Reliable Broadcast)
DaPoA
Nodes
─Reliable blockreplication across nodes
─Consistent block views
─Multiple block proposals & acceptance

21. 21
DaPoA: Operations
Tx
Req.TxPool
Consensus
round r-1round r
SubDAG
(Ordered Block
Digests)
with an index…
…
…
…
-Block
-round r
…
…
DAG Mempool
(Reliable Broadcast)
DaPoA
Nodes
─Consistent total order
─Local & Deterministic
─Based on a shared
randomness

22. 22
DaPoA: Operations
Tx
Req.TxPool
Consensus
round r-1round r
SubDAG
(Ordered Block
Digests)
with an index…
…
…
…
DAG Block
-Block
-round r Sub
DAG…
……
DAG Mempool
(Reliable Broadcast)
DaPoA
Nodes
Retrieves corresponding blocks

23. 23
DaPoA: Operations
Tx
Req.TxPool
Consensus
round r-1round r
SubDAG
(Ordered Block
Digests)
with an index…
…
…
…
Validate &
Commit
Validate &
Commit
Validate &
Commit
DAG Block
-Block
-round r Sub
DAG…
……
DAG Mempool
(Reliable Broadcast)
DaPoA
Nodes
Retrieves corresponding blocks

24. 24
Security Analysis
Round 1Round 2Round 3Round 4
2
1
3
4
5
Fork
Attack
Round 1Round 2Round 3Round 4
DAG MempoolConsensus
Accept a proposal with 2f+1 votes
from the next round

25. 25
Security Analysis
Round 1Round 2Round 3Round 4
2
1
3
4
5
Fork
Attack
Omission
Attack
Round advances despite f failures
DAG Mempool
Round 1Round 2Round 3Round 4
Consensus
Accept a proposal with 2f+1 votes
from the next round

26. 26
Security Analysis
Round 1Round 2Round 3Round 4
2
1
3
4
5
Round 1Round 2Round 3Round 4
Fork
Attack
Accept a proposal with 2f+1 votes
from the next round
Omission
Attack
Ordering based on
a shared randomness
Fairness
Attack
DAG MempoolConsensus
Round advances despite f failures

27. 27
Evaluation: vs. Clique
─Implement DaPoAbased on Ethereum Geth*
─Narwhal/Tusk** for DAG BFT
─Benchmark using Hyperledger Caliper
─Hardware: AMD Ryzen 3990X CPU, 256 GB RAM
Parameters
─Simple payment contract
─TX confirmation blocks: 2
─Block Period: 1
─Gaslimit: 109
─DAG BFT (Batch Size: 2MB, 1 worker, 1 primary)
─Send rate: 700
*https://github.com/ethereum/go-ethereum
**https://github.com/facebookresearch/narwhal

28. 28
Evaluation: vs. Clique
─Implement DaPoAbased on Ethereum Geth*
─Narwhal/Tusk** for DAG BFT
─Benchmark using Hyperledger Caliper
─Hardware: AMD Ryzen 3990X CPU, 256 GB RAM
Parameters
─Simple payment contract
─TX confirmation blocks: 2
─Block Period: 1
─Gaslimit: 109
─DAG BFT (Batch Size: 2MB, 1 worker, 1 primary)
─Send rate: 700
*https://github.com/ethereum/go-ethereum
**https://github.com/facebookresearch/narwhal
DaPoA scales better

29. 29
Evaluation: vs. Clique
─Implement DaPoAbased on Ethereum Geth*
─Narwhal/Tusk** for DAG BFT
─Benchmark using Hyperledger Caliper
─Hardware: AMD Ryzen 3990X CPU, 256 GB RAM
Parameters
─Simple payment contract
─TX confirmation blocks: 2
─Block Period: 1
─Gaslimit: 109
─DAG BFT (Batch Size: 2MB, 1 worker, 1 primary)
─Send rate: 700
*https://github.com/ethereum/go-ethereum
**https://github.com/facebookresearch/narwhal
5.76xlower
latency
DaPoA scales better
2.74xhigher
throughput

30. 30
DaPoA
lAn effort to enhance Ethereum Clique for performance scalability
─Multiple block acceptance
lLeveraging DAG-based BFT consensus
─A consistent block view across nodes using DAG mempool.
─A consistent block ordering using local consensus based on shared randomness
Conclusion