Adaptive Mechanisms for Large Scale Software Systems

本项目研究大型软件系统的自适应机制。它专注于模型，算法和架构，允许大规模软件系统自主响应环境和用户需求的变化。随着云计算服务的成熟和扩散，一个开放的研究问题是如何管理部署到分布式云的大规模软件系统，其中多个云基于地理位置、数据隐私、延迟或其他维度分布。 自适应软件系统是软件系统工程中的一个新的研究和实践领域，其目标是使软件更加通用、健壮、持续可用和自我管理。首先，使用分布式云系统的各种组合需要一个集成和分布式的自适应系统。 我们将探索安排反馈回路，自适应系统的基本构建块，在分层模式中优化低级别的本地云和更高级别的全局云。 我们将检查循环如何交互并收敛到全局解决方案，并确定如何最好地指定所需的全局解决方案。 其次，自适应系统可以根据系统条件、工作负载和操作环境以多种模式运行。 该目标解决了如何对这些多模式进行建模的问题，特别是需要哪些模型、环路和控制算法。 第三，为了提高决策质量和执行速度，我们将尝试叠加两个反馈循环，一个用于响应应用程序的长期趋势，另一个用于满足短期需求。 需要准确的趋势预测和劳动分工，以避免短期和长期目标之间的反复颠簸和冲突。 自适应系统中的这一重要工作使下一代云计算成为可能，特别是复杂系统的自动化，以降低成本，提高能源效率并提高服务质量。 这些贡献将推动加拿大工业、研究人员和公民非常感兴趣的实际应用领域的最新技术。

This project investigates adaptive mechanisms for large-scale software systems. It focuses on models, algorithms and architectures that allow large scale software systems to respond autonomously to changes in environment and user requirements. As cloud computing services mature and proliferate, an open research question is how to manage large scale software systems deployed to distributed clouds, where multiple clouds are distributed on the basis of geography, data privacy, latency, or other dimensions. Adaptive software systems are emerging as a new research and practice area in software systems engineering with the goal of making software more versatile, robust, continuously available and self-managing.First, using various combinations of distributed cloud systems requires an adaptive system that is both integrated and distributed. We will explore arranging feedback loops, the basic building blocks of adaptive systems, in hierarchical patterns to optimize both local clouds at the low level and globally at a higher level. We will examine how the loops interact and converge to a global solution, and determine how best to specify the desired global solution. Second, adaptive systems can function in multiple modes depending on the system conditions, the workload, and the operating environment. This objective addresses the question of how to model these multiple modes, and in particular what models, loops, and control algorithms are required. Third, to improve the quality of decisions and the speed of implementation, we will experiment with superimposing two feedback loops, one to respond to long term trends in the application and another to address short term needs. Accurate trend prediction and division of labor to avoid repeated thrashing and conflict between short and long term goals will be required. This important work in adaptive systems enables the next generation of cloud computing, in particular the automation of complex systems to decrease costs, increase energy efficiency, and improve the quality of service. These contributions will advance the state of the art in an area of practical application that is of great interest to Canadian industry, researchers, and citizens.