Replication data for: Multi-objective application placement in fog computing using graph neural network-based reinforcement learning

This dataset comprises a collection of synthetic application‐placement instance sets for heterogeneous cloud–edge/fog infrastructures, designed for the evaluation of single‐ and multi‐objective optimization strategies. Each instance describes:
- a directed acyclic graph (DAG) of interdependent services forming an application,
- a set of compute nodes (cloud, edge, fog) with resource capacities and connectivity latencies,
- resource demands of each service (e.g., CPU, memory), service‐to‐service dependency weights or communication cost,
- one or more placement solutions together with objective values (such as latency, energy consumption, deployment cost) generated by algorithms including the DRL model, a genetic algorithm (GA) and an NSGA-II multi‐objective heuristic.

The dataset is split into training and test sets and is generated via the provided instance_generator.py and generate_dataset.py scripts. It allows researchers to benchmark and compare placement algorithms in terms of Pareto-front coverage, convergence speed, and trade-offs between objectives.

Potential uses: Investigating learning‐based or heuristic algorithms for application placement, multi‐objective optimisation in the cloud/fog continuum, dependency‐aware placement of microservices, as well as enabling reproducibility and comparison across approaches.

本数据集包含一系列面向异构云-边/雾计算基础设施的合成应用部署实例集，旨在用于单目标与多目标优化策略的评估。每个实例涵盖以下内容：
- 构成应用的相互依赖服务的有向无环图（directed acyclic graph, DAG）
- 具备资源容量与连通延迟的计算节点（云、边、雾）集合
- 各服务的资源需求（如CPU、内存）、服务间依赖权重或通信开销
- 由深度强化学习（Deep Reinforcement Learning, DRL）模型、遗传算法（genetic algorithm, GA）以及NSGA-II多目标启发式算法等生成的一个或多个部署方案，及其对应的目标值（如延迟、能耗、部署成本）。

该数据集被划分为训练集与测试集，通过附带的instance_generator.py与generate_dataset.py脚本生成。研究人员可利用该数据集针对部署算法的帕累托前沿覆盖率、收敛速度以及目标间权衡等维度开展基准测试与对比研究。

本数据集的潜在应用场景包括：研究面向应用部署的基于学习或启发式算法、云-雾连续体中的多目标优化、感知依赖的微服务部署，以及助力各类研究方法的可复现性与跨方法对比。