The proposed Multi-Modal and Multi-Scale Temporal Propagation model can effectively support cloud-native AIOps and provide a solid technical foundation for proactive monitoring and fault diagnosis of microservice systems.
Abstract
To address the challenges of heterogeneous multi-source data, inadequate collaborative modeling of temporal and topological features, and low efficiency in dual-task optimization in performance prediction and bottleneck localization for cloud-native microservice systems, this paper proposes a Multi-Modal and Multi-Scale Temporal Propagation model (M[Formula: see text]TP). The model achieves unified representation of logs, time-series metrics, and call-chain data through a Multi-modal Heterogeneous Embedding Module, simultaneously captures dynamic evolutionary patterns and service topological dependencies via a Temporal-Graph Joint Learning Module, and implements joint optimization of performance prediction and bottleneck localization using a Dual-Task Collaborative Decoding mechanism. Experiments on two public datasets, GAIA and PetShop, demonstrate that the M[Formula: see text]TP model achieves [Formula: see text] scores of 0.95 and 0.96 for performance prediction, F1 scores of 0.93 and 0.94 for bottleneck localization, and inference latencies as low as 7.1ms and 6.5ms, respectively, outperforming 8 baseline models including LSTM, Informer, and GAT. Ablation studies validate the effectiveness of each core component, and case studies confirm the model's capability in capturing performance fluctuations and identifying root-cause services accurately. The proposed model can effectively support cloud-native AIOps and provide a solid technical foundation for proactive monitoring and fault diagnosis of microservice systems.
This study integrates three types of heterogeneous information to construct a unified embedding space, achieves cross-source alignment of semantically heterogeneous data through a hierarchical architecture, introduces differentiated dynamic weight allocation among three data sources, and employs multi-source context-guided sparse compensation to address missing entries in the service invocation matrix.
Zhenzhen Liu· Academic Journal of Computin...· 0 citations
AMF-CloudForge is presented, a unified machine learning-driven framework that integrates migration state analysis, intelligent scheduling, consistency preservation, and real-time adaptive management into a single end-to-end architecture and transforms cloud data migration from a static, tool-centric process into a reliable, adaptive, and continuously optimized cloud service.
S. Sapate, G. Pathak· Journal of Intelligent Decis...· 0 citations
: The dynamic variations in Quality of Service (QoS) within cloud computing environments pose significant challenges for accurate prediction. Addressing the issues of inadequate multi-source feature modeling and low prediction efficiency in temporal QoS prediction, this study integrates Long Short-Term Memory (LSTM) networks, Graph Attention Network (GAT), and attention mechanisms to develop a hybrid neural network modeling framework specifically designed for QoS prediction. The framework sequentially performs three key tasks: constructing temporal graph structures, integrating heterogeneous multi-source features, and enabling multi-task collaborative prediction, thereby effectively capturing the dynamic patterns of user preferences, network states, and service loads during service invocation. Experimental results demonstrate that the proposed framework outperforms existing baseline methods in both response time and throughput prediction tasks, and the multi-task learning strategy enhances prediction accuracy while significantly improving computational efficiency.
Zhenzhen Liu· Academic Journal of Engineer...· 0 citations
With the explosive development of LLM-empowered agent technology, LLM inference performance has become more important than training. Cloud computing is a popular deployment approach, where performance prediction is vital for instance selection and QoS assurance. However, prediction is challenging due to GPU hardware heterogeneity, Transformer operator variations, and dynamic inference configurations. Virtualization and other features vary across clouds, further increasing prediction difficulty. Existing methods suffer from low accuracy and poor generalization. To tackle these issues, we propose Dispeller, a prediction model for GPU-accelerated cloud environments with three feature sets: 1) basic GPU hardware feature with 7 dimensions; 2) operator-level GPU performance feature with 4 dimensions; 3) inference configuration feature with 4 dimensions. We conduct experiments on public cloud GPUs and collect a real-world dataset of 10,112 samples. Random Forest is adopted to learn the nonlinear mapping between features and performance. Experimental results show that Dispeller achieves high prediction accuracy with TPS $\mathrm{R}^{{2}} = 0.951$ on seen GPUs and 0.989 on unseen GPUs, demonstrating strong cross-GPU generalization. An ablation study confirms that inference configuration features contribute 87.3% of the predictive power. Dispeller is therefore able to recommend cloud resources and optimize LLM deployment costs.
Huan Zhou, Zhi-Peng Wang, Meng-Juan Li et al.· Fall Joint Computer Conferen...· 0 citations
Accurate workload prediction in cloud data centers is essential for efficient resource management, yet high-dimensional and noisy operational data often hinder forecasting performance. This work extends the original CVCBM model by integrating a lightweight Bidirectional GRU (BiGRU) with Bidirectional LSTM (BiLSTM) to enhance prediction efficiency while maintaining temporal feature extraction. Initially, workload signals are denoised and decomposed using a two-stage process—Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) followed by Variational Mode Decomposition (VMD). Sample Entropy (SE) selects meaningful components, and K-Means clustering prioritizes high workload data for training. The hybrid Conv1D-BiLSTM-BiGRU architecture captures multi-scale temporal patterns and both short-term and long-term dependencies. The trained model is deployed using the Flask framework for real-time workload prediction, allowing interactive input of datasets and immediate forecasting. Experimental evaluation demonstrates that the extended model reduces computational overhead while improving prediction accuracy, providing robust, scalable, and real-time forecasting for cloud data center resource management.
Rayala Ashok, M. Praveena, G. S. Prasad et al.· 2026 7th International Confe...· 0 citations
As smart manufacturing and networked sensing systems continue to evolve, distributed cloud platforms generate massive volumes of log data whose efficient analysis is essential for reliable monitoring and intelligent decision-making. Such capabilities also provide valuable support for communication-oriented and electromagnetic sensing infrastructures requiring real-time system awareness. To address the limitations of existing distributed log analysis methods, including insufficient feature extraction, weak capture of critical information, and the difficulty of balancing efficiency and accuracy, this paper proposes a distributed log analysis approach based on a Multi-Scale Convolutional Attention Mechanism (MS-CAM). A structured preprocessing pipeline is first established to perform log transformation and noise filtering. A multi-scale convolutional module is then employed to extract features at different granularities, capturing both local critical information and global semantic relationships, while an attention mechanism further enhances key feature representation through adaptive weight allocation. Experimental results demonstrate that the proposed method effectively improves analytical performance and provides an efficient solution for distributed intelligent systems with potential value for real-time monitoring and signal-aware computing applications.
X.-Y. Liu, D.-B. Luo, W.-J. Wang et al.· Advanced Electromagnetics· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
