标签化数据驱动的新型数据中心异构任务调度机制研究

The cloud data center is becoming the basic infrastructure in the modern information age, and it is also the necessary part in our ordinary lives. Hence, it has special practical significance to explore how to optimize the resource utilization, accelerate the task execution, and improve user experience, to satisfy the common users of data centers. As the rapid growth of data-intensive tasks, and the employment of NVM (Non-Volatile Memory), future data centers are facing significant changes on task object and storage architecture, which introduces more challenges, including eliminate the data resource bottleneck, coordinate heterogeneous storage devices, and maintain system stability. However, traditional computing-centric resource schedule methods are not efficient anymore since it lacks support for data management and consideration of heterogeneous storage devices. In this project, we focus on the heterogeneous task scheduling problem in future data centers, and we aim to imporve global task execution. We will design labeled cross expression model for task object and data object, reconstruct the relationship between tasks and data, evaluate network workload, and conduct active data migration. We will establish heterogeneous storage devices coordination based data update and replica management methods, explore labeled data driven differentiated task scheduling algorithms, and study multi-dimensional resource coordination based system recovery mechanism. Accordingly, we can optimize the task execution efficiency, and provide guaranteed quality of service.

数据中心正成为信息时代的社会基础设施，是社会生产生活不可或缺的一部分，探索如何优化数据中心资源利用、提升任务执行效率、改善用户体验，以满足更广泛的用户需求，具有切实的研究价值和意义。随着基于数据的服务不断涌现和新型非易失性内存的使用，未来数据中心将在任务对象和存储架构方面发生重大改变，并引发数据资源瓶颈破除、异构存储介质协同、系统稳定性保障等挑战性问题。以计算为中心的资源调度机制缺少对数据管理的支持，也未考虑异构存储介质的区别，不足以应对新型数据中心下异构任务调度面临的挑战。本项目针对新型数据中心下的异构任务调度问题，以优化任务执行效率为目标，设计任务与数据的标签化交叉表达模型，重构任务与数据之间的关系，评估网络负载，实施主动数据迁移，构建异构存储介质协同的数据更新与副本管理机制，探索标签化数据驱动的差异化任务调度机制，研究多维资源协同的系统修复机制，优化任务执行效率，保障系统服务质量。

项目以标签化数据驱动的新型数据中心异构任务调度机制为核心.利用新型非易失性存储介质的性能优势，形成了一套以标签化数据驱动的数据中心任务调度机制，开发了一套基于云平台环境的原型系统。基于项目研究成果，发表研究论文11篇，授权专利6项；相关研究内容在电力领域得到试点应用，项目执行期间获得2020年度江苏省科学技术奖二等奖。相关研究成果亦有望在国防领域得到进一步利用。