InstantChain: Enhancing Order-Execute Blockchain Systems for Latency-Sensitive Applications.

As an emerging replicated transactional processing system, blockchain is being adopted by more and more industries. Some blockchain-based applications require low latency and high throughput. However, the ordering phase is being a bottleneck of the order-execute blockchain. Poor ordering will result in low throughput due to transaction conflicts, and poor quality of service for latency-sensitive applications because their transactions are not processed within their expected times. Therefore, this paper enhances order-execute blockchain systems to minimize latency and maximize parallelism by considering transaction latency requirements, conflicts and varied execution times. First, we propose a parallelism-oriented and latency-sensitive transaction ordering method. Then, we propose a deep reinforcement learning-based method to determine optimal block parameters. Finally, we propose an adaptive packing method based on space occupancy that enables packed transactions to fully exploit parallelism. Experimental results show that our implemented system (InstantChain) outperforms 3 related systems under both real and simulated datasets.