新型体系结构下数据密集型计算的运行时优化机制研究

Data-itensive computing is very important now in many areas. But, how to provide a runtime susyem with high efficiency, eneygy saving and good reliability based on multi/many-cores environment is still a key challenge. As we all know, runtime system is a key part of system software to ensure system's security and high efficiency. Based on the multi/many-cores environment, the project will solve the problems for data-intensive computing. There are four key issues in the project, which are listed as follows: 1) resource monitoring, predicting, mapping and scheduling in runtime system; 2) performance adjustment and scheduling in runtime system; 3) energy measurement, predicting and energy-aware scheduling; 4) fault-tolerant models for threads/virtual machines. Based on the four research issues, we will implement the runtime system prototype for data-intensive super computing systems to verify the correctness of the above key proposed schemes. The success of the research will improve the running effects of data-intensive computing application on multi/many-cores environment.

数据密集型计算已经成为重要的计算类型，但基于多核/众核新型体系结构，如何为其构建高效、低耗、可靠的运行时支撑环境是一个重要的挑战。运行时系统是系统软件的重要组成部分，是确保系统安全与高效的关键环节。本课题以多核/众核等新型体系结构为硬件支撑，围绕数据密集型计算所存在的运行时执行效率不高、可靠性保障不足、能耗高等突出问题，以实现程序运行时的高效性、高可靠、低能耗为目标，对其展开深入研究。课题围绕着四个方面开展：1）运行时资源监控与预测、映射及管理模型；2）运行时性能优化与调度机制；3）运行时能耗度量、预测模型与能耗敏感的调度算法；4）运行时线程级/虚拟机级容错模型和可靠性保证机制。基于上述四个相互关联的研究内容，实现典型数据密集型计算的运行时原型系统，以验证所取得关键技术和算法。本课题的成功研究将有助于促进和改善数据密集型计算在多核/众核环境上的运行效果，推动多核技术的进步。

大数据处理已经是工业界和学术界共同关心的重要话题。如何为大数据处理提供一个高效率的处理环境，是一个重要的问题。本课题组针对多核/众核平台，提出了缓存在线划分与管理机制、能耗管理机制、OpenMP运行时优化机制、云游戏运行时优化机制等关键技术；部分成果在电力行业的数据处理领域有应用。总体上看，本项目共发表或录用论文30余篇，其中IEEE/ACM Ttrans.以及CCF A/B类会议论文17篇左右；申请了近10项国家发明专利；举办了包括ChinaSys、全国软件年会（NASAC）、CCF青年精英大会以及ICPADS 2016等国际、国内学术会议；培养了硕士、博士生共计20余名。