Areopagus: Asynchronous Sharding Blockchain Tolerating Corrupted Shards with Optimized Overhead
Abstract
Sharding technology alleviates the scalability bottleneck of traditional blockchains by partitioning the network into multiple shards, thereby improving throughput through parallel consensus execution. However, limited fault tolerance within individual shards degrades overall system fault tolerance and necessitates large shard size, ultimately constraining performance. In this paper, we present Areopagus, which tolerates shard corruption in asynchronous networks, enabling smaller shard sizes and improved performance. Areopagus introduces a decoupled and abridged transaction processing pattern that ensures secure intra-shard and cross-shard transaction processing with corrupted shards. Corrupted shards that pose risks of inconsistent commitment or liveness loss are guaranteed to be detected and recovered, while malicious nodes can be identified. Furthermore, Areopagus employs a corruption-isolated finalization mechanism to achieve responsive transaction finalization, regardless of whether the corresponding shard is corrupted. Intra-shard transactions are finalized with corrupted shard tolerance at a cost comparable to standard consensus. Cross-shard transaction processing requires at most a single agreement instance, thereby reducing overhead compared with state-of-the-art solution (Camael, CCS'25), which relies on multiple agreement executions. Extensive experiments conducted on 1000 AWS EC2 nodes across 4 regions demonstrate that Areopagus achieves a peak throughput of 738.58 ktx/sec, improving throughput by 7.4 times over Camael and by 23.9 times over the baseline (Kronos, NDSS'25).