Dynamic Runtime Software Architecture Adaptation

随着社会计算基础设施的普及，不断产生、获取和分析数据的方式发生了范式转变。据报道，世界上90%的数据是在过去两年中产生的。预测表明，未来几年，数据本身的增长率将是指数级的。前所未有的数据量增长给处理如此海量数据的软件平台带来了令人兴奋的新挑战。从软件开发的角度来看，软件应用程序需要表现出可扩展性、弹性和可靠性等理想特性，以满足与数据量和数据增长率相关的最低要求。虽然社区已经认识到对创新软件平台的迫切需求，但越来越多的解决方案没有解决软件体系结构层的可伸缩性、灵活性和可靠性问题，而是通过数据分区、数据复制和分发模型(如MapReduceTM)等技术将这些重要的考虑因素转移到数据管理层；因此，数据管理问题与应用程序的可伸缩性和灵活性纠缠在一起。这一拟议的软件平台运行时重新配置和架构适配研究方案将开发架构模型、算法、技术和工具，以有效地处理和解决与数据的新特征相关的挑战，如数据量、多样性、速度和准确性。主要目标是开发动态的运行时可重构和自适应的软件体系结构模型，该模型将自动反应和适应不同的数据生产和消费模式。我们在该研究计划中工作的具体目标将是：a)通过开发模型验证和确认技术来实现实时体系结构适应，所述模型验证和确认技术无需预先设定规则即可自动评估体系结构备选方案；b)通过对指向体系结构级应用程序划分或合并需求的采样的低开销应用简档数据执行分析来提供可伸缩性和弹性；c)支持软件体系结构在运行时经济高效地回滚到已经被简档和稳定的体系结构检查点，以便主动地防止或重新主动地解决不期望的质量或功能状态；以及d)通过对软件应用程序的历史行为与可测量的软件体系结构指标相关联的建模，实现预测性自适应可靠性评估。拟议的计划为HQP培训提供了坚实的基础，包括2名博士和4名硕士研究生，他们将获得最先进的软件工程、自适应系统和大数据应用方面的专业知识。

There has been a paradigm shift in the way data is constantly produced, procured and analyzed specially with the prevalence of social computing infrastructure. It has already been reported that 90% of all the data in the world has been generated over the last two years. Predictions indicate that the data growth rate itself will be exponential in the next few years. The unprecedented growth of data volumes has introduced exciting new challenges for the software platforms that deal with such voluminous amounts of data. From a software development perspective, software applications need to exhibit desirable characteristics such as scalability, elasticity, and reliability in order to meet the minimum requirements associated with data volume and data growth rates. While the community has acknowledged the pressing need for innovative software platforms, a growing number of solutions do not address issues of scalability, elasticity and reliability at the software architecture layer, offloading these important considerations onto the data management layer through techniques such as data partitioning, data replication and distribution models such as MapReduce; therefore, entangling issues of data management with application scalability and elasticity. This proposed research program on runtime reconfiguration and architectural adaptation of software platforms will develop architectural models, algorithms, techniques and tools that will efficiently handle and address challenges associated with the novel characteristics of data such as volume, variety, velocity and veracity. The main objective will be to develop dynamic runtime reconfigurable and adaptive software architectural models that would automatically react and adapt to varying data production and consumption patterns.The concrete objectives of our work in this research program will be to: a) enable real-time architecture adaptation by developing model verification and validation techniques that automatically evaluate architectural alternatives without prior rule setting; b) provide scalability and elasticity by performing analysis on sampled low-overhead application profiling data that would point to architecture-level application partitioning or merging requirements; c) support runtime cost-efficient rollback of software architecture into already profiled and stable architecture checkpoints in order to proactively prevent or reactively resolve an undesirable quality or functional state; and d) enable predictive adaptation reliability estimation through modeling historical behavior of software applications as they correlate with measurable software architecture metrics.The proposed program provides a solid foundation for HQP training including 2 PhD and 4 MASc students who will gain expertise in state-of-the-art software engineering, self-adaptive systems, and big data applications.