GBShard: Adaptive Sharding for Blockchain Network via Granular-Ball Partitioning

Abstract

Blockchain scalability remains constrained by cross-shard transaction overhead and workload imbalance in conventional sharding architectures. To address these limitations, we propose GBShard, an adaptive sharding method grounded in granular-ball computing theory. The method partitions blockchain transactions to shards through iterative splitting and merging of coarse-grained granular-ball structures, enabling adaptive shard formation with minimized cross-shard transactions. A multi-granularity dynamic sharding algorithm further optimizes topology by redistributing granular-ball groups in dynamic scenarios, achieving incremental sharding adjustment while reducing the overhead of global-repartitioning. Experimental validation using real-world Ethereum transaction traces demonstrates GBShard's superiority over Monoxide-based sharding schemes: it achieves 1.3–1.8× higher throughput, 34–42% lower transaction latency, and reduces cross-shard transactions by 23–50%. These results suggest granular-ball partitioning as a viable strategy for adaptive blockchain scaling.