Synchronization of transaction pools (mempools) has shown potential for improving the performance and block propagation delay of state-of-the-art blockchains. Indeed, various heuristics have been proposed in the literature to this end, all of which incorporate exchanges of unconfirmed transactions into their block propagation protocol. In this work, we take a different approach, maintaining transaction synchronization outside (and independently) of the block propagation channel. In the process, we formalize the synchronization problem within a graph theoretic framework and introduce a novel algorithm (SREP - Set Reconciliation-Enhanced Propagation) with quantifiable guarantees. We analyze the algorithm's performance for various realistic network topologies, and show that it converges on any connected graph in a number of steps that is bounded by the diameter of the graph. We confirm our analytical findings through extensive simulations that include comparison with MempoolSync, a recent approach from the literature. Our simulations show that SREP incurs reasonable overall bandwidth overhead and, unlike MempoolSync, scales gracefully with the size of the network.

翻译：交易池（mempool）的同步在提升先进区块链性能与区块传播延迟方面展现出潜力。事实上，已有多种启发式方法被提出用于此目的，这些方法均将未确认交易的交换纳入其区块传播协议中。本研究采取不同策略，将交易同步置于区块传播信道之外（独立进行）。在此过程中，我们基于图论框架形式化定义了同步问题，并提出了一种具有可量化保证的新型算法（SREP——集合协调增强传播）。我们分析了该算法在多种现实网络拓扑下的性能，证明了该算法在任意连通图中收敛所需步数受图直径限制。通过涵盖与最新方法MempoolSync对比的大规模仿真，我们验证了分析结果。仿真表明，SREP带来合理的总体带宽开销，且与MempoolSync不同，其性能随网络规模扩展而优雅化。