Skip to content
Blockworks

The Optimal Tradeoffs for Protocol Design | Day 1 | Protocol R&D

Nikhil Chaturvedi, Wei Dai, Alex Watts, Patrick O'Grady

0:000:00

Key takeaways

  • User-Centric Finality: All speakers converge on the view that raw block time matters less than how quickly users feel confident a transaction will settle; designs must optimise for perceived, not paper, latency.
  • Verticalisation Wave: Patrick O'Grady predicts the most successful apps will own more of the stack, using techniques like application-specific sequencing to customise ordering, fees and UX.
  • Asymmetry Is Coming: Wei Dai argues that “generation-three” chains will embrace heterogeneous nodes, letting heavyweight data-centre validators coexist with home-stakers and ZK provers to scale capacity dynamically.
  • DSMR + Multi-Proposer Paths: Both Patrick O'Grady and Alex Watts see decoupled state-machine replication plus multiple concurrent proposers as the cleanest route to high throughput without a proposer monopoly.
  • Networking, not Consensus, is the Bottleneck: The panel stresses that P2P layers such as GossipSub are now the drag on performance; projects like Turbine and Optimum P2P that price and prioritise bandwidth will be decisive.
  • Pro-Spam, Anti-DoS: Allowing cheap failed transactions is viewed as a feature because it lets chains charge for work and stay DoS-resistant, though networks must rate-limit at the transport layer to protect liveness.
  • Looming Governance Risks: Vendor lock-in around wallets, ERC-7702 and bridges, plus difficult privacy trade-offs, are emerging non-technical threats that could fragment users across incompatible silos.

On Next-Gen Chain Architecture

Alex Watts would start with Monad’s high-throughput, EVM-compatible engine but fix its single-proposer “monopoly” by allowing several proposers to run in parallel. Wei Dai separates today’s “generation-two” speed races from a coming “generation-three” that welcomes node asymmetry: specialised data-centre validators supply bulk compute while light nodes and home-stakers preserve censorship resistance. Patrick O'Grady champions decoupled state-machine replication, then sells block bandwidth to dedicated sequencers; users send a transaction to the nearest data centre and receive one-round-trip confirmation, while the chain later references only certificates, not raw data. All three designs treat ultrafast execution and flexible sequencing as table stakes rather than differentiators.

On Latency, Throughput and UX

Watts would choose a one-second block packed with 100 million gas over a half-second block holding 40 million gas, because extra compute lets contracts absorb complexity—gasless trades, escrowed liquidity, richer settlement—so users need not wait for explicit confirmation. O'Grady reframes the metric as “user-centric finality”: the moment a customer believes funds are safe. That perception may come from synchronous inclusion, escrow guarantees or asynchronous proofs, but networks that shave milliseconds while starving compute will lose to those that make interactions feel instant.

On Application-Specific Sequencing

The trio regard application-specific sequencers as the key experiment of 2024. FastLane Labs is betting that decentralised applications will want private control of ordering to optimise MEV capture, fee markets and spam filtering without abandoning shared liquidity. O'Grady cautions that fully breaking out into a solo chain often strands liquidity and composability; his bet is that good asynchronous bridges plus vertical control of sequencing can give apps both autonomy and connection. The jury is still out on where the boundary between shared and siloed execution should lie.

On MEV Spam and Fee Dynamics

Flashbots data show Base blocks now contain heavy on-chain solver logic that succeeds only one percent of the time. Wei Dai notes that despite the noise, median fees fell because supply of block space grew faster than spam demand—proof that scaling still helps users. Watts and O'Grady defend “failed” transactions as a DoS safety valve: charging for wasted execution deters attackers more cheaply than off-chain filtering. The real hazard is saturating network ports, a lesson from Solana’s past congestion; intelligent rate limiting and stake-weighted quality-of-service are preferred over moral policing of contract logic.

On Networking Bottlenecks

Consensus engines can already hit sub-second confirmation, but propagating blocks and attestations remains slow. GossipSub ignores real-world topology and floods peers indiscriminately. Solana’s Turbine and Optimum P2P propose tree-based or coded multicast that acknowledges link capacity and lets validators express which data they will accept. The panel expects overlay networking to become a shared service layer, much like CDNs for Web 2, giving every chain low-latency, high-bandwidth transport.

On Future Fault Lines

State, Privacy, Vendor Lock-in

Key actionable insights

  • Track adoption of multi-proposer designs and DSMR implementations as early indicators of a post-single-leader paradigm.
  • Monitor development in overlay networking projects that promise bandwidth pricing and prioritisation; they may become mandatory for high-throughput rollups.
  • Watch for industry pushback or regulation around wallet and bridge vendor lock-in, a non-technical risk that could shape user acquisition and retention.