基于信息-能源-时间三元组的算力资源综合度量与优化方法研究

摘要：现有算力度量方法以浮点运算次数（FLOPS）为单一指标，忽视了能效与时间效率，导致资源调度不合理。为此，提出一种基于“信息-能源-时间“三元组的算力资源综合度量与优化方法。构建了以最优硬件为基准的归一化评分模型：S = w₁×(I/Imax) + w₂×(η/ηmax) + w₃×(Tmin/T)，将算力、能效比、时间3个维度归一化至[0,1]区间，并支持性能优先、节能优先、响应优先等场景化权重配置。基于“算网大脑“的仿真验证表明，该方法能够有效提升资源利用率，降低能耗并优化任务执行时间，为算力网络的精细化资源管理提供了新方案。

关键词：算力网络；算力度量；信息-能源-时间三元组；归一化评分；能效比；动态调度

Abstract: Existing computing power metrics rely solely on Floating Point Operations Per Second (FLOPS) as a performance indicator, neglecting energy efficiency and time efficiency, which leads to suboptimal resource scheduling. To address this issue, a comprehensive measurement and optimization method for computing power resources based on the “information-energy-time” triplet are proposed. A normalized scoring model benchmarked against optimal hardware is constructed as S = w₁×(I/Imax) + w₂×(η/ηmax) + w₃×(Tmin/T), which normalizes the three dimensions of computing power, energy efficiency ratio, and time into the [0,1] range, and supports scenario-specific weight configurations such as performance priority, energy-saving priority, and response priority. Simulation verification based on the “Computing Network Brain” shows that the proposed method can effectively improve resource utilization, reduce energy consumption, and optimize task execution time, providing a new solution for fine-grained resource management in computing power networks.

Keywords: computing power network; computing power measurement; information-energy-time triplet; normalized scoring; energy efficiency ratio; dynamic scheduling