Smart Contract Parallel Execution with Fine-Grained State Accesses

As various optimizations being proposed recently, the performance of blockchains is no longer limited by the consensus protocols, successfully scaling to thousands of transactions per second. To further improve blockchains' throughput, exploiting the parallelism in smart contract executions becomes a clear solution to resolve the new performance bottleneck. The existing techniques perform concurrency control on smart contract transactions based on pre-determined read/write sets, which can hardly be calculated precisely. As a result, many parallelization opportunities are missed in order to maintain the correctness of transaction executions. In this paper, we propose a novel execution scheduling framework, DMVCC, to further increase the parallelism in smart contract executions, via more fine-grained control on state accesses. DMVCC improves over existing techniques with two key features: (1) write versioning, eliminating the write-write conflicts between transactions, and (2) early-write visibility, enabling other transactions to read the writes from a transaction earlier, before it being committed. We integrated DMVCC into the Ethereum Virtual Machine, to evaluate its performance in real-world blockchain environments. The experimental results show that DMVCC doubles the parallel speedup achievable to a 20x overall speedup, compared with the serial execution baseline, approaching the theoretical optimum.