面向智能合约的并行随机超级优化及运行时窥孔优化方法

The performance problems of smart contracts will waste the resources of blockchain, transaction fees, and decelerate the processing of transactions. The research on the optimization of smart contracts has shortcomings in automatic degree, the alleviation of search space explosion and the optimization of deployed smart contracts. Applicant’s previous studies show that the resources consumption of smart contracts is not properly measured, and peephole optimization can improve the performance of smart contracts. Based on applicant’s previous studies, this project plans to research superoptimization and peephole optimization for smart contracts with full consideration of blockchain’s features. This project consists of four research contents. First, we propose context-aware measurement based on the analysis of execution traces of smart contracts. During the measurement, we will compute the execution probability of program paths and measure the resources consumption incurred by running the atomic operations of smart contracts in fine granularity, which supports the subsequent optimization. Second, we will design a parallel stochastic superoptimization algorithm based on MapReduce, to accelerate the construction of peephole optimization rules. Third, we will try to improve the load balance of workers based on the intelligent algorithm. Our approach takes in runtime feedback of workers and is expected to increase resources utilization. Fourth, we will conduct the research on Just-in-Time (JIT) peephole optimization to overcome the challenge that the deployed smart contracts cannot be modified, and investigate the way of reducing its runtime overhead based on network evolvement analysis. This project will improve the performance of smart contracts and reduce resources consumption of blockchain. Besides, this project can facilitate the design of blockchain protocol infrastructure and virtual machine instruction set.

智能合约的性能问题会浪费区块链资源、交易费用以及降低交易处理速度。智能合约优化的研究在自动化程度、缓解搜索空间爆炸、优化已部署合约等方面存在不足。申请人前期研究发现：智能合约的资源消耗未被准确测量；窥孔优化能够提高智能合约的性能。鉴于此，本项目充分考虑区块链特性，研究智能合约超级优化及窥孔优化方法，包括四个内容。1) 提出上下文感知的测量方法，深入分析智能合约执行轨迹，计算程序路径执行概率，实现对原子操作资源消耗的细粒度测量，为优化提供支持。2) 设计基于MapReduce的并行随机超级优化方法，加速构建窥孔优化规则。3) 结合节点运行时反馈，基于智能算法改善计算节点负载均衡，提高资源利用率。4) 开展运行时窥孔优化研究，解决已部署合约无法修改的问题，并基于网络演化分析降低优化过程的时间开销。本课题的研究将提高智能合约性能，降低区块链资源消耗；对区块链协议架构和虚拟机指令集设计有指导意义。

本项目拟先探索如何测量智能合约原子操作对计算资源的消耗；随后设计并行随机超级优化(stochastic superoptimization)算法，自动化的生成窥孔优化规则；最后设计轻量级的原子操作运行时窥孔优化方法。如何进行程序优化，提高程序性能一直都是程序语言、软件工程相关领域孜孜以求的目标。在本文中，我们提出了一种新颖的解决方案，利用以太坊虚拟机 (EVM) 处理函数的方式来自动恢复函数签名。在区块链领域，程序优化还具有特别的意义，因为有性能问题的智能合约将会浪费节点计算资源，同时会导致交易处理速度变慢，区块链吞吐量降低乃至收取过多的交易费，从而极大的影响用户体验。因此，研究智能合约优化方法具有重要意义。