HATLedger: An Approach to Hybrid Account and Transaction Partitioning for Sharded Permissioned Blockchains

With the development of sharded blockchains, high cross-shard rates and load imbalance have emerged as major challenges. Account partitioning based on hashing and real-time load faces the issue of high cross-shard rates. Account partitioning based on historical transaction graphs is effective in reducing cross-shard rates but suffers from load imbalance and limited adaptability to dynamic workloads. Meanwhile, because of the coupling between consensus and execution, a target shard must receive both the partitioned transactions and the partitioned accounts before initiating consensus and execution. However, we observe that transaction partitioning and subsequent consensus do not require actual account data but only need to determine the relative partition order between shards. Therefore, we propose a novel sharded blockchain, called HATLedger, based on Hybrid Account and Transaction partitioning. First, HATLedger proposes building a future transaction graph to detect upcoming hotspot accounts and making more precise account partitioning to reduce transaction cross-shard rates. In the event of an impending overload, the source shard employs simulated partition transactions to specify the partition order across multiple target shards, thereby rapidly partitioning the pending transactions. The target shards can reach consensus on received transactions without waiting for account data. The source shard subsequently sends the account data to the corresponding target shards in the order specified by the previously simulated partition transactions. Based on real transaction history from Ethereum, we conducted extensive sharding scalability experiments. By maintaining low cross-shard rates and a relatively balanced load distribution, HATLedger achieves throughput improvements of 2.2x, 1.9x, and 1.8x over SharPer, Shard Scheduler, and TxAllo, respectively, significantly enhancing efficiency and scalability.