Jun 2026· Cluster Computing· Vol 29· 0 citations· 40 references
Computer Science
TL;DR
This paper proposes a Machine Learning-based framework designed to predict key performance indicators, including CPU utilization, memory, and energy consumption, based on incoming workload patterns, while simultaneously forecasting potential system overload conditions, and introduces a profiling methodology that characterizes serverless functions according to their resource consumption profiles.
Abstract
The Function as a Service (FaaS) paradigm has emerged as a compelling architectural model for both cloud and edge computing environments, enabling the execution of self-contained functions triggered by specific events while abstracting from developers infrastructure management complexities such as load balancing and auto-scaling. In FaaS-enabled clusters, particularly within resource-constrained edge environments, precise resource consumption estimation becomes critical to optimize resource utilization, minimize latency, prevent system overloads, and ensure scalability. This paper addresses performance modeling challenges in FaaS-enabled distributed and decentralized edge computing systems, operating at the granularity level of both nodes and individual functions. We propose a Machine Learning-based framework designed to predict key performance indicators, including CPU utilization, memory, and energy consumption, based on incoming workload patterns, while simultaneously forecasting potential system overload conditions. Moreover, our approach introduces a profiling methodology that characterizes serverless functions according to their resource consumption profiles, thereby enabling accurate prediction of node-level resource demands without requiring detailed knowledge of individual deployed functions. Experimental validation demonstrates that our predictive models achieve 97% accuracy in anticipating node overload scenarios, providing a robust foundation for proactive resource management in edge-based FaaS deployments. Moreover, while our best individual, function-based regression models predict node-level CPU, RAM and power consumption with a Mean Absolute Percentage Error below 9% on average, our experiments highlight the effectiveness of function profiling and cluster-based modeling. On one hand a novel multi-target regressor based on a permutation-invariant neural architecture is proved to generalize effectively across previously unseen workload compositions in the tested scenarios, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R^2$$\end{document} scores ranging from 0.94 to 0.98, i.e., aligned with the performance of specialized single-target regression models. On the other hand, cluster-level models generalize effectively to previously unseen functions within the same usage class, maintaining prediction errors within practically acceptable ranges and often in the single-digit percentage range.
One of the major challenges encountered by cloud computing environments is the ability to handle dynamic workloads at a minimal energy usage and quality of service. This paper discusses a smart resource optimization system, which uses machine learning to forecast resource needs and assign virtual machines to the most efficient tasks. Through a proactive strategy, over-provisioning and under-provisioning that are traps of manual or rule-based scaling are reduced by the system. The method is concentrated on the combination of predictive modelling and automated scaling triggers. The studies apply the CloudSim Plus simulation toolkit as an environment modelling tool and Google Cluster Data traces as the main dataset. In particular, the investigation runs a portion of these traces to recreate heterogeneous real-world task requirements. The outcomes show that energy spending and rejection rates of tasks are significantly lowered as compared to traditional methods of allocation which are not dynamic. The metrics used in the performance are CPU utilization and memory overhead. As in the paper, machine learning models can dramatically improve the operational performance of the large-scale cloud infrastructures by making real-time data-driven decisions, which forms a solid basis behind the next-generation autonomous cloud management systems.
Resource-constrained edge devices require efficient and adaptive health monitoring to ensure reliable operation under dynamic workloads. This paper proposes an LLM-driven, context-aware framework that integrates real-time system metrics, historical data, and task-specific importance levels for anomaly detection and prediction. Specifically, the framework forecasts the semantic health state of the edge device five minutes ahead based on recent monitoring observations, enabling proactive intervention before critical operating conditions are reached. Multidimensional metrics, including CPU, memory, temperature, network load, and process information, are transformed into structured time-series representations and used as input to pre-trained machine learning models. A large language model (LLM) acts as an orchestration layer, dynamically selecting the most appropriate predictive model based on system context and resource constraints. This enables adaptive trade-offs between accuracy, latency, and computational cost. Experimental results on Raspberry Pi devices show that the proposed approach achieves comparable or improved performance while reducing resource usage compared to static methods.
Ioannis Tzitzios, A. Dimara, Georgiana Petridou et al.· Electronics· 0 citations
CELLServe formalizes SLO-constrained joint resource provisioning as an optimization problem with a dedicated algorithm, and introduces an opportunistic instance merging strategy for decode phase functions to reclaim fragmented resources.
Zejian Wang, Nan Lin, Zinuo Cai et al.· ACM Transactions on Architec...· 0 citations
Managing containerized workloads in cloud-native infrastructures poses complex challenges due to the need to simultaneously balance performance, efficiency, and sustainability. This work proposes an adaptive resource allocation framework that leverages Digital Twins for real-time system monitoring and integrates Large Language Models to support context-aware decision-making under multi-objective constraints. The proposed approach dynamically optimizes latency, bandwidth utilization, and energy consumption, enabling intelligent workload orchestration across heterogeneous data center environments. A flexible utility function is introduced to allow system operators to adjust trade-offs between responsiveness and environmental impact. Experimental results demonstrate that the framework consistently outperforms traditional heuristic and learning-based baselines, achieving higher allocation accuracy, improved network utilization, and faster workload completion, while reducing overall energy consumption by more than 20% in sustainability-oriented scenarios. These findings highlight the potential of combining digital twins-driven observability with large language model-based reasoning to enable interpretable, adaptive, and energy-efficient resource management in next-generation cloud computing environments.
Pedro Henrique Sachete Garcia, A. Lorenzon, M. Luizelli et al.· SN Computer Science· 0 citations
The growing demand for AI-driven workloads, particularly from Large Language Models (LLMs), has raised concerns about the significant energy and resource consumption in data centers. This work introduces a novel LLM-based predictive scheduling system designed to enhance operational efficiency while reducing the environmental impact of data centers. Our system utilizes an LLM to predict key metrics such as execution time and energy consumption from source code, and it has the potential to extend to other sustainability-focused metrics like water usage for cooling and carbon emissions, provided the data center can track such data. The predictive model is followed by a real-time scheduling algorithm that allocates GPU resources, aiming to improve sustainability by optimizing both energy consumption and queuing delays. With fast inference times, the ability to generalize across diverse task types, and minimal data requirements for training, our approach offers a practical solution for data center scheduling. This framework demonstrates strong potential for advancing sustainability objectives in AI-driven infrastructure. Through our collaboration with a data center, we achieved a 32% reduction in energy consumption and a 30% decrease in waiting time.
Hanzhao Wang, Jingxuan Wu, Yumeng Li et al.· 0 citations
The new EMC+ proposal is an OS‐driven elasticity manager for container‐based environments that continuously estimates idle core cycles left by regular (inelastic) applications, and reallocates idle cores to elastic ones, even during short time intervals, and has minimal impact on the performance and QoS of colocated inelastic applications.
J. C. Saez, Carlos Bilbao, Manuel Prieto-Matías· Concurrency and Computation· 0 citations