Mecury: Integrating Account Allocation and Scheduling for Enhancing Blockchain Sharding Performance

Abstract

Existing sharded blockchain systems suffer from excessive cross-shard communication overhead and workload imbalance. Although state-of-the-art account allocation schemes partially alleviate these problems, they still fail to simultaneously achieve effective cross-shard transaction reduction and balanced workload distribution, while also lacking adaptability to dynamic workloads. In particular, current methods lack mechanisms for efficient incremental allocation of newly joined accounts and cannot dynamically adjust workloads during surges, resulting in performance degradation until the next account allocation occurs. To address these limitations, we propose Mecury, the first scheme to realize account allocation and scheduling that span entire blockchain epochs. We propose a Fine-Grained Label Propagation Algorithm combined with a Workload-Aware Incremental Allocation Algorithm to achieve high-quality allocation of both historical and newly joined accounts. Furthermore, Mecury introduces an account proxy mechanism equipped with a dynamic proxy algorithm that reduces cross-shard transactions by establishing temporary proxy relationships, thereby adaptively mitigating workload surges. In addition, we design a secure cross-shard state migration protocol to ensure the liveness and consistency of state migration between accounts involved in the proxy relationship. By integrating these mechanisms, Mecury effectively addresses the challenges of excessive cross-shard transactions and workload imbalance inherent in sharded blockchains during consensus. Experimental evaluations on real Ethereum data demonstrate that Mecury consistently delivers superior performance, achieving up to 72% higher throughput and 53% lower latency compared to state-of-the-art baselines.