Smart Distributed Software and Systems

当前和新出现的许多类型的复杂系统-包括大数据系统、云数据基础设施、物联网和网络物理系统-具有越来越多的分布式和动态架构，在创建和支持应用程序方面提供了前所未有的灵活性。然而，这种高度分布式的架构也增加了管理性能、配置、维护和安全性的复杂性。软件定义环境的出现和机器智能的最新进展使得分布式软件系统能够自我管理，同时符合最终用户的目标。拟议的研究推动了自主计算的理论，实践和应用的前沿，使自适应系统足够智能，以保证其在现实世界中的正常运行。 自治管理系统（AMS）是自适应系统中提供自管理功能的组件。我们将利用白盒建模（即，排队和控制理论）和黑盒建模（即，机器学习模型）来为目标分布式系统构建高效可靠的AMS。白盒模型捕捉系统的本质，而黑盒模型估计白盒模型无法捕捉的运行时不确定性和未知参数。虽然这种混合方法是一个有前途的研究方向，可能导致自适应系统在真实的世界中的广泛使用，但需要同时解决以下挑战： 1)大型分布式系统需要分散管理，这就需要多个AMS之间的协调。解决多个AMS之间的冲突和定义它们之间的关系已经成为这一领域的明显挑战。 2)到目前为止，AMS一直专注于变革管理，而不是目标管理。目标可能会因持续的人为驱动的开发、集成和交付过程（也称为软件演化）而发生变化，这需要纳入AMS的体系结构中。 3)虽然最佳配置无法实现，但可接受的配置是可行的，但实现起来具有挑战性。有必要设计AMS，使可接受的配置和部署成为任何分布式系统的现实。 总之，这项研究计划1）有助于自主计算的显着进步，2）培训熟练的HQP与合法和及时的专业知识和3）铺平了道路，广泛使用自适应系统在健康，能源和城市管理在加拿大和国外。这种智能自适应系统同时提高了上述部门提供的服务的效率、准确性、可靠性和可用性。因此，该研究计划不仅会影响学术研究，还会影响最终导致社会改善的行业。

Current and emerging complex systems of many types--including big data systems, cloud data infrastructure, Internet of Things and cyber-physical systems have increasingly distributed and dynamic architecture that provide unprecedented flexibility in creating and supporting applications. However, such highly distributed architectures also increase the complexity of managing performance, configuration, maintenance and security. The emergence of software-defined environments and recent progress in machine intelligence make it possible to enable distributed software systems to self-manage while conforming to the end-users' objectives. The proposed research pushes the frontiers of the theory, practice, and applications of autonomic computing to make self-adaptive systems smart enough to guarantee their proper functioning in real-world scenarios. The autonomic management system (AMS) is the component that provides self-management in self-adaptive systems. We will leverage white-box modeling (i.e., queuing and control theories) and black-box modeling (i.e., machine learning models) to build efficient and reliable AMSs for target distributed systems. The white-box models capture the essence of the system while the black-box models estimate the runtime uncertainties and unknown parameters that white-box models fail to capture. Although this hybrid approach is a promising research direction that may lead to widespread use of self-adaptive systems in real world, the following challenges need to be addressed in parallel: 1) Large-scale distributed systems require decentralized management which raises the need for coordination among multiple AMSs. Resolving the conflicts among multiple AMSs and defining relationships among them have emerged as evident challenges in this area. 2) So far, AMSs have been concentrated on change management rather than goal management. Goals may change by the continuous human-driven process of development, integration and delivery, aka software evolution, which needs to be incorporated in the architecture of AMSs. 3) While optimal configuration is not attainable, acceptable configuration is feasible but challenging to achieve. There is a need to design AMSs that can make acceptable configurations and deployment a reality for any distributed system. In summary, this research program 1) contributes a notable advancement in autonomic computing, 2) trains skilful HQPs with rightful and timely expertise and 3) paves the path for widespread use of self-adaptive systems in management of health, energy and cities here in Canada and abroad. Such smart self-adaptive systems improve the efficiency, accuracy, reliability and availability of services offered in the above-mentioned sectors simultaneously. Therefore, this research program will impact not only academic research but also industries that ultimately lead to the betterment of society.