Imitater: An Efficient Shared Mempool Protocol with Application to Byzantine Fault Tolerance

TL;DR

The paper addresses efficiency and availability challenges in Byzantine Fault Tolerant consensus for blockchains by introducing Imitater, a Shared Mempool protocol that decouples transaction distribution from consensus and guarantees totality through microblock chaining and erasure coding in a setting with $n=3f+1$ replicas and up to $f$ Byzantine nodes. It develops a two-phase SMP (Dispersal and Retrieval) that uses per-node microblock chains, availability certificates, and Merkle proofs to ensure all honest nodes eventually obtain microblock content, enabling order-preserving decision making. Imitater is integrated into HotStuff as Imitater-HS and into Fast HotStuff as Imitater-FHS, with formal safety and liveness arguments and a comprehensive performance analysis. Large-scale experiments on clusters with up to 256 nodes demonstrate significant throughput improvements and lower latency compared with Stratus-based baselines, confirming practical scalability under faulty conditions.

Abstract

Byzantine Fault Tolerant (BFT) consensus, a cornerstone of blockchain technology, has seen significant advancements. While existing BFT protocols ensure security guarantees, they often suffer from efficiency challenges, particularly under conditions of network instability or malicious exploitation of system mechanisms. We propose a novel Shared Mempool (SMP) protocol, named Imitater, which can be seamlessly integrated into BFT protocols. By chaining microblocks and applying coding techniques, Imitater efficiently achieves \emph{totality} and \emph{availability}. Furthermore, a BFT protocol augmented with Imitater ensures \emph{order preservation} of client transactions while mitigating the risks of \emph{over-distribution} and \emph{unbalanced workload}. In the experiment, we integrate Imitater into the HotStuff protocol, resulting in Imitater-HS. The performance of Imitater-HS is validated in a system with up to 256 nodes. Experimental results demonstrate the efficiency of our approach: Imitater-HS achieves higher throughput and lower latency in the presence of faulty nodes compared to Stratus-HS, the state-of-the-art protocol. Notably, the throughput improvement increases with the number of faulty nodes.

Figures (5)

• Figure 1: Chain-based organization of a node's microblock mempool. Each node maintains $n$ local microblock chains, where the $i$-th chain sequentially records microblocks disseminated by node $i$.
• Figure 2: The communication pattern of disseminating a microblock in SMP
• Figure 3: Throuput with $n$=49 and bandwidth settiing to 100Mbps, varying faulty nodes number $f$
• Figure 4: Performance comparison of Imitater-HS and Stratus-HS with bandwidth setting to 100 Mbps and up to $n/3$ nodes being faulty. (a) Throughput vs. nodes. (b) Latency vs. nodes.
• Figure 5: $B$ and $B^*$ both getting committed (impossible).

Theorems & Definitions (22)

• lemma thmcounterlemma: Dispersal Termination
• proof
• lemma thmcounterlemma: Microblock Uniqueness
• proof
• theorem thmcountertheorem: SMP Totality
• proof
• theorem thmcountertheorem: SMP Chain-Consistency
• proof
• theorem thmcountertheorem
• theorem thmcountertheorem