RECAP Artificial Data Traces

The objective of the work package "Data Collection, Visualization and Analysis" of RECAP is to provide the necessary tools for managing and refining the data needed for the rest of the work packages. This includes the collection as well as the generation of data. Within this work package, the task of Artificial Workload Generation is responsible for the generation of a collection of datasets with artificial workloads, that complement the real data traces collected from industrial partners. Moreover, because publicly available workload data is scarce we provide the data as public data sets. This document is a companion report to deliverable which is of type “dataset”. The aim of the report is to describe the collection of datasets that constitute D5.3 and the mathematical techniques (structural time series models, generative adversarial networks, and workload based on traffic propagation) by which one can artificially generate and/or augment such datasets. The datasets described include real data traces collected by industrial partners and artificial data traces generated by the use of statistical models and neural networks. Each published data set can be used by the scientific and industrial community as a starting point for the modelling and experimental validation of distributed edge and cloud applications, facilitating the repeatability of the results.RECAP项目“数据收集、可视化与分析”工作包的目标，是为其余工作包所需的数据管理与优化提供必要工具，涵盖数据的收集与生成两大环节。在该工作包中，“人工工作负载生成”任务负责生成一批搭载人工工作负载的数据集集合，作为从工业合作伙伴处采集的真实数据轨迹的补充内容。此外，鉴于公开可用的工作负载数据较为稀缺，我们将这批数据以公开数据集的形式对外发布。本文档为对应“数据集”类型交付物的配套报告，旨在阐述构成D5.3的数据集集合，以及可用于人工生成或扩充此类数据集的三类数学技术：结构化时间序列模型、生成式对抗网络，以及基于流量传播的工作负载生成方法。本次描述的数据集包含两部分：一是工业合作伙伴采集的真实数据轨迹，二是通过统计模型与神经网络生成的人工数据轨迹。每一份公开发布的数据集，均可作为科学与工业界开展分布式边缘与云应用建模、实验验证的起始基础，助力相关研究结果的可复现性。