Content area
Concurrency control is crucial for ensuring consistency and isolation in distributed transaction processing. Traditional concurrency control algorithms, such as locking-based protocols, usually suffer from performance degradation due to heavy transaction coordination overheads. To overcome this problem, deterministic concurrency control approaches are widely adopted in many systems since they can avoid coordination overhead by eliminating uncertainty. In these systems, every node receives identical transaction batches, orders them according to specific rules, and executes them concurrently in a determined correct sequence. However, some transactions might have to be aborted in concurrent execution, wasting expensive network bandwidth and computing resources. We find that this problem significantly lowers system performance, especially in geographically distributed settings where network communication is a bottleneck. To exploit deterministic concurrency control efficiently in geo-distributed application scenarios, this paper studies an optimized deterministic concurrency control approach GB-DCC for permissioned blockchain applications which is a new type of distributed transaction processing systems. Three general optimization strategies are proposed: deterministic pre-execution, mini-batch partitioning, and deterministic re-execution. Experiments show that under the YCSB-A benchmark workload, these strategies can reduce the distributed system’s bandwidth consumption by 17.8% and improve the performance obviously.