Minimmit: Fast Finality with Even Faster Blocks

Achieving low-latency consensus in geographically distributed systems remains a key challenge for blockchain and distributed database applications. To this end, there has been significant recent interest in State-Machine-Replication (SMR) protocols that achieve 2-round finality under the assumption that $5f+1\leq n$, where $n$ is the number of processors and $f$ bounds the number of processors that may exhibit Byzantine faults. In these protocols, instructions are organised into views, each led by a different designated leader, and 2-round finality means that a leader’s proposal can be finalised after just a single round of voting, meaning two rounds overall (one round for the proposal and one for voting). We introduce Minimmit, a Byzantine-fault-tolerant SMR protocol with lower latency than previous 2-round finality approaches. Our key insight is that view progression and transaction finality can operate on different quorum thresholds without compromising safety or liveness. Experiments simulating a globally distributed network of 50 processors, uniformly assigned across ten virtual regions, show that the approach leads to a 23.1% reduction in view latency and a 10.7% reduction in transaction latency compared to the state-of-the-art.

💡 Research Summary

Minimmit is a Byzantine‑fault‑tolerant state‑machine‑replication (SMR) protocol that reduces latency beyond existing 2‑round finality schemes by decoupling the quorum thresholds used for view progression and for block finalisation. Under the stronger resilience assumption 5f + 1 ≤ n (where n is the total number of processors and f the maximum number of Byzantine faults), Minimmit requires only 2f + 1 signed votes (an “M‑notarisation”) to move to the next view, while still demanding n − f votes (an “L‑notarisation”) to consider a block finalised. This separation exploits the fact that in globally distributed networks a small quorum can be collected much faster than a large one, thereby shortening the time between views without sacrificing safety.

The paper provides formal proofs of consistency, liveness, and optimistic responsiveness (the guarantee that a correct leader after GST can finalise a block in a single round of voting). The key safety argument shows that if a block receives an L‑notarisation, no other block in the same view can ever obtain 2f + 1 votes, because the two vote sets would intersect in at least f + 1 processors, guaranteeing at least one correct processor would have voted twice – a contradiction. Consequently, any processor that sees an M‑notarisation can safely start the next view, knowing that the block it voted for cannot be contradicted later.

Experimentally, the authors built a network simulator that reproduces inter‑region latency and jitter measured from AWS over the past year. They placed 50 processors uniformly across ten virtual regions (US‑west‑1, US‑east‑1, EU‑west‑1, AP‑northeast‑1, etc.) and compared Minimmit against Simplex (a 3‑round protocol) and Kudzu (a 2‑round protocol that uses the larger quorum n − 2f for view changes). Results show Minimmit achieving a view latency of 146 ms (σ = 21 ms), a 23 %–25 % reduction relative to Simplex (194 ms) and Kudzu (190 ms). Block latency (time to finalise a block) is 220 ms, matching Kudzu and improving on Simplex’s 299 ms. Because a transaction submitted just after block height h must wait for the next view and the subsequent block’s finalisation, the reduced view latency translates into a transaction end‑to‑end latency of 366 ms (146 ms + 220 ms) versus 410 ms for the other 2‑round protocols—a 10.7 % improvement.

The protocol deliberately drops the “slow path” (an extra voting round used when the fast path fails) to avoid the latency penalty of waiting for a larger quorum. This design choice slightly reduces resilience to crash faults compared with protocols such as Alpenglow or Hydrangea, which retain a slow path and can tolerate additional crash failures. However, under the 5f + 1 ≤ n assumption, Minimmit still guarantees safety and liveness as long as at most f processors are Byzantine.

Beyond the core algorithm, the authors describe several engineering optimisations. They integrate aggregate and threshold signatures to bound communication during asynchronous periods, and they present an erasure‑coded variant (E‑Minimmit) that expands block data by a factor of 2.5 but reduces per‑processor bandwidth. These techniques are generic and could be applied to other “Simplex‑style” protocols.

In comparison with related work, Minimmit’s main contribution is the insight that view progression can safely use a much smaller quorum than finalisation, yielding measurable latency gains in realistic, geographically dispersed deployments. The paper positions Minimmit as a practical solution for high‑throughput, low‑latency blockchain or distributed database systems where the stronger fault‑tolerance assumption is acceptable. Future work is suggested on extending the approach to weaker resilience bounds, adaptive quorum selection, and deployment on real‑world testbeds.