To address the issues of limited network size， strong dependency on network environment， high storage cost， and low transaction throughput existing in the scalability of blockchain systems， a sharding mechanism adapted to asynchronous network environments and supporting parallel transaction processing was proposed. In this mechanism， HoneyBadgerBFT consensus was used to achieve data consistency in asynchronous networks， sharding technology was employed for linear scalability of blockchain system， and DAG （Directed Acyclic Graph） technology was adopted to further enhance capability for intra-shard and inter-shard parallel transaction processing. Simulation results demonstrate that the proposed mechanism maintains liveness even in the asynchronous network environment； in the semi-synchronous network environment， the proposed mechanism reduces the communication overhead by more than 49.9% compared to SharPer using PBFT （Practical Byzantine Fault Tolerance）； in a blockchain network composed of 16 nodes， the proposed mechanism has the TPS （Transactions-Per-Second） decreased by 16.7% compared to SharPer， while in a 64-node blockchain network， the proposed mechanism has the TPS higher than SharPer by 6.7%， showing higher throughput of the proposed mechanism than SharPer； under the condition of 20% cross-shard transactions and using the same network environment and hardware resources， when the numbers of shards and nodes double， transaction throughput of the proposed mechanism grows 30.0% and 10.5% respectively more than that of SharPer， demonstrating better scalability of the proposed mechanism than SharPer.