Tarp-data

The picture depicts our proposed adaptive resource allocation approach based on graph neural networks for optimizing qos-aware interactive microservices in cloud computing. This method uses DAG topology to extract the global characteristics of microservices, and adaptively generates microservice resource allocation strategies, which can effectively use microservice resources while ensuring the quality of service. This method uses EGAT to extract microservice features and uses reinforcement learning to generate resource allocation policies.First, we define the microservice state graph.Then, we use EGAT to generate embeddings for each node in the graph by extracting the hidden features of resources and network metrics.Based on the message pass paradigm of graph neural networks (GNN), we design the microservice feature passing to capture correlations between microservices, thereby improving the transferability of our approach. Finally, we use DDPG to model microservices in a uniform and self-adaptive manner. 本图示展示了我们提出的基于图神经网络的自适应资源分配方法，该方法用于优化云计算环境中服务质量(QoS)感知的交互式微服务。该方法采用有向无环图(DAG)拓扑结构提取微服务的全局特征，并自适应生成微服务资源分配策略，可在保障服务质量的前提下高效利用微服务资源。本方法采用EGAT提取微服务特征，并结合强化学习生成资源分配策略。首先，我们定义了微服务状态图。随后，通过提取资源与网络指标的隐藏特征，我们利用EGAT为图中的每个节点生成嵌入表征。基于图神经网络(GNN)的消息传递范式，我们设计了微服务特征传递机制以捕捉微服务间的关联，进而提升本方法的可迁移性。最后，我们采用深度确定性策略梯度(DDPG)以统一自适应的方式对微服务进行建模。

搜集汇总

背景与挑战

背景概述

Tarp-data数据集聚焦于云计算中基于图神经网络的微服务资源分配优化，采用DAG拓扑和EGAT方法提取全局特征，结合强化学习自适应生成策略，旨在提升服务质量和资源效率。