Efficient Execution of Blockchain Transactions Through Deterministic Concurrency Control

Abstract

Concurrently executing Blockchain transactions can make good use of modern hardware and improve system performance. Recent works mainly utilize a dependency graph to represent the partial order among conflicting transactions. All participants can then execute the transactions concurrently according to an identical order and keep consistent with each other. However generating dependency graphs is quite time-consuming, and during this process some transactions have to be re-executed multiple times. In this paper, we adopt deterministic concurrency control to quickly compute the partial order. Instead of one large graph, we partition the transactions into a sequence of batches. In this way, there is no need to deal with the inter-batch conflicts in finding the partial order. We also propose a two stage approach (DVC) to find a partial order with high degree of parallelism. DVC does not need to re-execute transactions, and can find an approximate optimal partial order by solving an equivalent MinVWC problem. We integrate the proposed techniques into an open-source system and compare them with several advanced solutions. As shown by the experimental results, our approaches can significantly reduce the costs of computing the partial order and obtain a schedule of high parallelism degree.