通信竞争异构系统中基于逆向分层的能耗与可靠性联合优化策略及方法

High performance and high throughput have been the two main goals pursued by heterogeneous computing systems, and the reliability of the system is the key to the running success of the parallel application. Due to the objective reality of communication link contention in heterogeneous computing systems, this study introduces shared recovery and DVFS technologies to explore the low energy consumption and high reliability theory of parallel tasks in heterogeneous computing systems on the basis of the idea of upward rank. As low energy consumption and high reliability for computing systems are conflicting, an energy-efficient optimization scheduling algorithm for communication awareness of parallel task under deadline constraints will be devised in this study. Considering the reliability of the communication link and the instantaneous fault of the processor brought by the DVFS technology in the energy saving process, a novel reliability enhancement algorithm of parallel tasks under the energy budget constraint is to be explored. Based on the heterogeneous computing system with communication contention, a dual-objective optimization algorithm with a lower failure rate and a smaller energy consumption is to be designed in the process of executing the task set with precedence constraints. Finally, some real parallel applications are used to verify and improve the developed parallel algorithms on JMetal, CloudSim, GridSim and other platforms. New ideas for the parallel task scheduling with low energy consumption and high reliability in heterogeneous computing system will be provided. In addition, this results can also be used for reference to other parallel applications.

高性能和高吞吐量是异构计算系统追求的两个主要目标，而系统的可靠性更是关乎并行应用能否运行成功的关键。鉴于异构计算系统中通信链路竞争的客观现实，本项目基于逆向分层的思想，引入共享恢复和DVFS技术，探索并行任务集在异构计算系统中的低能耗和高可靠性理论。由于低能耗与高可靠性相冲突，将开展截止时间约束下的并行任务集通信感知的低能耗优化算法研究。考虑通信链路的可靠性和DVFS技术在节能过程中产生的处理器瞬时故障，探索能量预算约束下并行任务集可靠性加强算法。基于存在通信竞争的异构计算系统，进一步设计有优先约束关系的任务集在执行的过程中产生较低的故障率和消耗较小的能量的双目标优化算法。最后，针对真实的并行应用，基于JMetal、CloudSim、GridSim等平台，验证和完善所设计的并行算法。研究成果将为并行任务在通信感知的异构系统中的低能耗和高可靠性优化供新思路，还可资其他并行应用问题借鉴。

项目资助期间，立足于异构分布式计算，结合共享恢复和DVFS技术，基于逆向分层细想，侧重于新理论和新方法的研究，采用理论与实践相结合，围绕并行任务集的低功耗和高可靠性两个主要目标，完成了截止时间约束下并行任务集通信感知的低能耗优化算法设计；基于通信竞争下的异构系统，设计了能耗约束下并行任务集可靠性加强策略，提出了并行任务集低能耗和高可靠性的双目标优化方法；发表学术论文13篇，其中IEEE Transactions论文1篇，SCI论文3篇，EI期刊论文3篇，CSCD论文7篇，申请国家发明专利1项。