分布异构计算资源通过网络连接形成算力网络(Computing power network，CPN)，其以"连"和"算"为核心。针对广分布异构性导致可行解空间巨大、强不确定性导致可行解空间易变、高约束复杂性导致可行解孤岛繁多、多目标性导致冲突目标权衡优化难等挑战，提出一个多层次算力网络体系框架，包括参数化结构化业务管理、三阶段(计划、调度、执行)闭环调度模式、多模态资源管理三个功能。提出支持快速、高效、鲁棒的"算法+知识+数据+算力"的算力网络智慧调度框架，形式化分析可行解空间，解析调度策略关键参数，定性分析调度算法性能与效率的内在关系，详细综述调度算法类型，综述算力网络调度研究进展与发展方向。对比已有相关综述研究，展望算力网络调度未来理论和技术的难点与趋势。
Survey on Wise Scheduling in Computing Power Network
Distributed heterogeneous computing resources are connected through the network to form a computing power network(CPN),in which"connection"and"computing"are the cores.There are several great challenges in-cluding the huge feasible solution space resulted from widely distributed heterogeneity,the variable feasible solu-tion space caused by strong uncertainty,a large number of feasible solution islands resulted from highly con-strained complexity,and the difficulty in optimizing multiple conflicting objectives.For these challenges,a multi-level computing power network framework is proposed in this paper,which contains three functions:Parameterized structural process management,three-phase(planning,scheduling,execution)closed-loop scheduling model,and multi-modal resource management.A computing power network intelligent scheduling framework with"algorithm+knowledge+data+computing power"is proposed to wisely schedule complex tasks to CPN resources effectively,efficiently and robustly.The feasible solution space is formally analyzed.The key parameters of scheduling strategy are introduced.The intrinsic relationship between effectiveness and efficiency of scheduling algorithms are qualitat-ively analyzed.Different types of scheduling algorithms are surveyed thoroughly.The state-of-the-art and future work are illustrated.The surveys on CPN scheduling algorithms are compared,followed by the theoretical and tech-nical problems and tendency in CPN scheduling algorithms.
Computing power network(CPN)cloud computingedge computingresource schedulingknowledge
国家科技期刊平台
NSTL
万方数据
