多模态网络算网效率对比数据集

本数据集面向算力网络（算网）融合场景下的效率评估与算法对比研究需求建设。数据集基于多模态网络资源协同调度系统在实际运行大模型业务与点播业务过程中产生的实测数据构建，系统记录了在协同与非协同两种资源调度算法下，网络拓扑中各节点的关键性能指标，包括平均负载均衡度、业务平均处理时延、业务达成率等，并提供了两种算法下的效率对比结果。数据内容涵盖从系统开始运行至结束的全程监控，以高频率（每3-5秒）采集并生成了包括平均时延、归一化Q值、成功率、负载均衡度等多项核心指标的时间序列数据。数据集实体文件包括avDelay.xlsx、successRate.xlsx、allAverageLoad.xlsx和CNE.xlsx四个主要数据表，分别详细记录了不同业务类型下的时延表现、任务完成情况、系统负载状态以及综合算网效率（CNE）计算结果。此外，数据集还附有第三方测试报告，分别验证了效率对比功能的正确性以及多模态资源调度算法在协同模式下的性能提升效果。适用于科研人员、工程师在算网协同调度、资源优化、性能评估等领域的定量分析与算法验证，为提升算网一体化系统的资源利用效率和任务处理能力提供重要数据支撑。数据量916KB。This dataset is developed to meet the research needs of efficiency evaluation and algorithm comparison in the computing power network (computing-network, CN) convergence scenario. The dataset is built based on measured data generated by a multimodal network resource collaborative scheduling system during the actual operation of large language model (LLM) services and video-on-demand (VOD) services. The system records key performance indicators (KPIs) of each node in the network topology under two resource scheduling algorithms: collaborative and non-collaborative, including average load balancing degree, average service processing delay, service completion rate, etc., and provides efficiency comparison results between the two algorithms. The data covers full-process monitoring from the start to the end of the system operation, and time-series data of multiple core indicators including average delay, normalized Q-value, success rate, load balancing degree are collected and generated at a high frequency (every 3-5 seconds). The physical dataset files include four main data tables: avDelay.xlsx, successRate.xlsx, allAverageLoad.xlsx and CNE.xlsx, which respectively record in detail the delay performance, task completion status, system load status and comprehensive computing-network efficiency (CNE) calculation results under different service types. In addition, the dataset is attached with third-party test reports, which verify the correctness of the efficiency comparison function and the performance improvement effect of the multimodal resource scheduling algorithm in the collaborative mode respectively. It is suitable for researchers and engineers to conduct quantitative analysis and algorithm verification in the fields of computing-network collaborative scheduling, resource optimization, performance evaluation, etc., providing important data support for improving the resource utilization efficiency and task processing capability of the integrated computing-network system. The total data size is 916 KB.

搜集汇总

背景与挑战

背景概述

该数据集面向算力网络融合场景，用于效率评估与算法对比研究，基于多模态网络资源协同调度系统在实际运行中的实测数据构建，包含协同与非协同两种资源调度算法下的关键性能指标对比。数据以高频率采集生成时间序列，涵盖时延、成功率、负载状态和综合算网效率等核心指标，并附有第三方测试报告，适用于科研和工程领域的定量分析与算法验证。