MEV mitigation using modular DAG-based mempools

Exploring MEV in the modular blockchain stack - John Adler (Celestia)

Modularity all the way down

Two approaches to MEV

Problem with Threshold decryption & Ordering fairness

New solution using DAG-based mempools

Narwhal: DAG-based mempool

Decouple mempool from consensus

—---------------------------------> network speed throughput

More than just research

Narwhal-TD: Threshold Decryption in the workers

See our repo for more info

Architecture

Next steps: Adding Time-Based Order-Fairness

Real Goal: integrating it into an actual blockchain

Mir-Themis: batch-order fairness with high “goodput”

Sending Tx to all validators

Providing high throughput

Not dividing bandwidth �by O(n)

?

Necessary Narwhal Modifications

​

• DAG transport must reference 3F+1 preceding messages (rather than 2F+1).
• Transactions are hashed and workers are assigned a hash space :
• If tx(hash) = Worker(hashspace), transaction data is batched.
• If tx(hash) =/= Worker(hashspace), transaction is kept in-memory only hash is batched.
• Ordering : Graph of tx dependencies is computed by primaries at commit time using batch unspooling technique – each vertices ordering completes the graph in a deterministic way.
• Hashed transactions are then retrieved in-memory or requested to any other worker.

​

Conclusion

Future ideas

• Dag mempool property: decentralized block building (everyone is a block proposer, every round)
• Bullshark needs neither a view-change nor a view-synchronization mechanism!
• Analogy with PBS to explore
• PBS on workers/primaries