Work

Multi-VM-Workloads Learning-based Forecasting Model for Resilient Server Management in Industry Clouds

Year: 2025

Type: conference-paper

Abstract: Forecasting server contention in large-scale cloud data center networks is crucial for efficient cloud resource management. The existing research primarily focuses on single-task machine learning methods for predicting resource-constrained VM failures, which often results in lower accuracy and higher computational costs. To overcome these challenges, this paper introduces a novel approach aimed at enhancing resilient server management in industry clouds. The proposed approach develops Multi-tasks-Learning-based Long Short-Term Memory for multiple VMs workload forecasting model named MVL-LSTM to simultaneously learn resource requirements of all VM hosted on a common server. It facilitates a single-step forecasting of workload resource requirements for all VM segments hosted on a server within a cluster. Experimental results using real-world Google Cluster VM datasets demonstrate that MVL-LSTM significantly improves forecasting accuracy while achieving faster execution and higher energy efficiency compared to existing single-task machine learning algorithms. Specifically, MVL-LSTM model achieves a 15% reduction in processing time and a 50% decrease in energy consumption compared with LSTM-based approach.

Source: Procedia Computer Science

DOI: 10.1016/j.procs.2025.03.275

Authors: Deepika Saxena, Ashutosh Kumar Singh

Institutions: University of Aizu, Indian Institute of Information Technology Bhopal

Language: English

Topic: Cloud Computing and Resource Management

Subfield: Information Systems

Field: Computer Science

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: diamond

Funder: University of Aizu