VTGAN based proactive VM consolidation in cloud data centers using value and trend approaches

Area of Science:

• Cloud Computing
• Resource Management
• Machine Learning

Background:

• Cloud data centers face challenges in energy consumption and resource optimization.
• Current virtual machine consolidation strategies rely on immediate workload fluctuations, leading to inefficiencies and service-level agreement (SLA) breaches.
• Existing time series and machine learning models struggle to predict dynamic cloud workloads accurately.

Purpose of the Study:

• To introduce a novel consolidation strategy for cloud data centers using predictive models.
• To enhance resource management by accurately forecasting future workloads and their trends.
• To reduce energy consumption, performance degradation, and SLA violations.

Main Methods:

• Development of hybrid value trend generative adversarial network (VTGAN) models to predict resource utilization and workload trends.
• Simulation of the proposed VTGAN approaches using real PlanetLab workloads on Cloudsim.
• Comparison of VTGAN strategies (value and trend-based) against baseline algorithms (THR-MMT-PBFD).

Main Results:

• The VTGAN (Up current and predicted trends) approach reduced SLA violations by 79% and virtual machine migrations by 56% compared to THR-MMT-PBFD.
• Integrating VTGAN into the VM placement algorithm to avoid predicted overloaded hosts further improved performance.
• Excluding predicted overloaded servers reduced SLA violations by 84% and VM migrations by 76% compared to THR-MMT-PBFD.

Conclusions:

• Predictive workload and trend analysis using VTGAN models offers a superior approach to cloud resource management.
• The proposed strategy significantly enhances data center efficiency by minimizing unnecessary migrations and SLA breaches.
• VTGAN models provide a robust solution for optimizing cloud resource utilization and energy consumption.