Aparna Krishnan is head of education at Blockchain at Berkeley and co-founder of Mechanism Labs, an open source blockchain research lab. Earlier this year, Aparna was awarded a scholarship by the DFINITY Foundation for Mechanism Labs’ research into consensus mechanisms. This episode is essentially a primer for advanced discussion of consensus in decentralized networks.
Aparna Krishnan Co-founder mechanism labs open source research lab All work and research is on Github Telegram: @mechanism_labs Co-founder of the education team at Blockchain Berkeley Consensus researcher Teaches executive education courses
Consensus Proof of Work Proof of Stake Old Field
History Cynthia Dwork developed stronger adversarial models Did not have many applications Blockchain has brought cryptography and consensus into the mainstream eye
Protocols DFINITY Tendermint Bitcoin Focus has been on proof of stake protocols Mining may not be sustainable Long term sustainabilities and lack of externalities is important Proof of stake offers efficiently As does proof of elapsed time and proof of space and time
Consensus Sybil control Coming to agreement relies on traditional consensus PoS, PoW refer to the sybil control mechanism PoS - Putting down capital Financial penalty for misbehavior Token holders are participants PoW - burning energy One cpu, one vote No connection between token custody and rewards P o Elapsed Time Proof of Authority placing reputation A cost of playing ball Traditional Consensus (PBFT) No concept of probabilistic finality All honest nodes come to final agreement Closed, permissioned Blockchain All nodes may with a high degree of probability agree A probabilistic guarantee
Longest Chain Rule Longest sequence of blocks is the Ethereum Ghost Can be attacked by a stealthy entity Open, permissionless, decentralized Inefficient Node churn Better liveness properties
Early PoS 2013 PeerCoin NXT Bitshares Primitive State grinding attack vulnerable Randomness derived from blockhashes
New Generation DFINITY Uses threshold relay decentralized randomness Tendermint Round robin Has social layer fallback Ethereum Casper Thunderella Algorand Hides the block creator until the block has been created
In Tendermint a minority below the assumption cannot break finality In Bitcoin >10% can change the finality through threatening to censor
Hard to get both economic and cryptographic security. Not clear that bribing and collusion attacks have been solved. No protocols have shown sophisticated economic models.
Mechanism Labs Focussing on incentive schemes What does it mean to have a stable, scalable protocol Scalability
Proof of Replication Recent work with Verifiable delay functions In commit/reveal schemes there is an opportunity to manipulate by giving only one person the ability to reveal the randomness Verifiable delay functions allow anyone to reveal the randomness ASIC resistance because not parallelizable