进化优化与知识共融的并行程序低耗测试理论与方法

The reliability of parallel programs has been a focus in both academic and industrial communities due to their widespread applications. Testing is a feasible way to guarantee the reliability. However, the computational consumption for testing parallel programs is often too high to be acceptable. Aiming at the adequate criterion of path coverage, the project studies evolutionary optimization and knowledge based theories and methods of testing message-passing parallel programs in low computational consumption, and expects to obtain the following achievements: (1) presenting a method of choosing target paths based on Latin hypercube sampling, so as to reduce the number of paths to be covered, (2) proposing an approach to evaluating schedule sequences according to the similarity between paths, based on which only schedule sequences with good quality are selected to execute the program under test, so as to reduce the number of running the program, (3) developing a strategy of generating test data based on evolutionary optimization, in which coarse estimation and precious calculation are employed to cost-efficiently obtain the fitness of individuals, and (4) validating the effectiveness of the proposed theories and methods by applying them to test benchmark or complex message-passing parallel programs in real-world applications. The proposed theories and methods can greatly reduce the computational consumption for testing parallel programs, significantly extend the scopes of testing complex parallel programs, and establish a technical foundation for green software testing. The project, at the cross edge of several disciplines, is very novel and challenging, and is urgently demanded by our country. Therefore, the project is of considerable significance in theory and application.

并行程序的广泛应用，使得其可靠性成为学术与产业界共同关注的焦点。通过测试能够提高并行程序的可靠性，但是，测试所需的计算消耗往往很高。本项目研究进化优化与知识共融的消息传递并行程序路径覆盖低耗测试理论与方法。通过研究，期望提出基于拉丁超立方采样的目标路径选择方法，减少需要覆盖的目标路径；提出基于路径相似度的调度序列性能评价方法，通过选择高性能的调度序列运行程序，减少程序运行的次数；提出进化个体适应值估计与计算共融的测试数据进化生成方法，通过估计进化个体的适应值，减少个体适应值的计算消耗；通过在基准和实际的复杂消息传递并行程序的应用，验证所提理论与方法的有效性。提出的理论与方法，能够大幅度降低测试所需的计算消耗，显著扩大在实际的复杂并行程序测试的应用范围，为软件的绿色测试奠定技术基础。这是多学科有机交叉、十分新颖且富有挑战性的研究方向，并与国家需求非常契合，具有重要的理论意义和实用价值。

并行程序的广泛应用，使得其可靠性成为学术与产业界共同关注的焦点。通过测试能够提高并行程序的可靠性，但是，测试所需的计算消耗往往很高。项目负责人及成员对进化优化与知识共融的消息传递并行程序路径覆盖低耗测试理论与方法，进行了为期4年的全面深入研究，提出了消息传递并行程序路径覆盖低耗测试理论与方法，大幅度降低了测试所需的计算消耗，为软件的绿色测试奠定了技术基础；实现了提出的理论与方法，并应用于大量实际的复杂消息传递并行程序测试中，为消息传递并行程序的低耗测试提供了应用范例。基于上述成果，获山东省高等学校科学技术一等奖1项、智能优化与调度学术会议优秀博士学位论文1篇和优秀论文二等奖1项、授权发明专利2项、计算机软件著作权登记2件，在被SCI或EI等检索的国际国内学术期刊上发表论文17篇，其中，SCI源刊论文14篇（含国际顶级期刊论文8篇），培养博士研究生5名、硕士研究生4名，圆满完成了项目的预期目标。研究成果解决了并行程序路径覆盖测试数据自动生成所需计算消耗高的难题，丰富了软件结构覆盖测试理论，提高了并行程序测试效率，扩大了进化优化的应用范围，具有重要的理论意义和实用价值。