UnionChain: Graph-based Shard Partitioning for Scalable Smart Contract Execution

Abstract

Sharding is a key technology for enhancing blockchain scalability by splitting the network into multiple shards, enabling parallel transaction (TX) processing. However, frequent cross-shard TXs can significantly limit the performance of sharding. Existing approaches reduce cross-shard TXs by modeling account relationships as a graph, but these methods only focus on efficiently processing normal transfer TXs between two parties. In practice, most blockchain TXs are generated by smart contracts, which involve multiple parties simultaneously, thus increasing graph complexity and prolonging shard partitioning time. In this paper, we propose UnionChain, a novel sharding protocol that reduces cross-shard TXs while efficiently handling smart contract TXs. Specifically, UnionChain first models the relationship between accounts and smart contracts as a weighted graph. By applying a community detection algorithm to partition this graph, it groups closely related entities into the same shard, thereby reducing cross-shard TXs and supporting efficient smart contract execution. To further address the graph complexity issue, UnionChain introduces an efficient vertex merging mechanism. When TXs occur between smart contracts, the corresponding contract vertices are merged into a single vertex. This mechanism significantly reduces graph size and shortens partitioning time. We implement a UnionChain prototype and evaluate its performance using a real Ethereum TX dataset. Compared to state-of-the-art sharding protocols, UnionChain shortens the partitioning time by up to 27% while maintaining a low cross-shard TX ratio, demonstrating its effectiveness in improving blockchain scalability.