The Latency Cost Of Censorship Resistance
Ling Ren
Session summary
The presenter studies the efficiency cost of providing strong censorship resistance in leader-based blockchain consensus. They model block production with rotating leaders and define censorship resistance as requiring that any sufficiently high-fee transaction submitted before a deadline must appear in the first block produced after its submission, whenever a leader actually proposes a block. This strengthens the usual liveness guarantee, which allows adversarial leaders to exclude a transaction for many blocks. Latency is measured in communication rounds between submission and finalization, focusing on “good case” latency when the leader is honest and the network behaves well.
The authors present a simple leader-based protocol that adds one communication round: clients send transactions to validators, validators forward them to the leader, and validators only vote for blocks that include the transactions they forwarded (subject to fee-based capacity constraints). This mechanism lets validators detect selective exclusion. The authors prove lower bounds showing that any censorship-resistant consensus protocol of this type must use at least four rounds in synchronous networks and five rounds in partially synchronous networks with higher fault tolerance, even in the good case, implying an inherent one-round overhead versus non-censorship-resistant designs.
Key takeaways
- The authors formalize censorship resistance as requiring that any sufficiently high-fee transaction submitted before a deadline be included in the first block produced after submission, whenever a leader chooses to propose a block.
- They propose a leader-based protocol where clients send transactions to validators, validators forward them to the leader, and validators refuse to vote for blocks that omit the transactions they forwarded, enabling detection of censorship attempts.
- Their lower bounds show that, in this leader-based model, censorship resistance necessarily increases good-case consensus latency by one round, to at least four rounds in synchrony and five rounds in partial synchrony with higher fault tolerance.