Optimizing Cloud Cloud Resource Allocation and Power Management: Attentive-Mamba and Reinforcement Learning Approaches

The high investment cost with low resource utilization has been a major concern for cloud providers. The efficient operation of clusters and achieving stable and accurate task load predictions are crucial. We introduce Attentive-Mamba, a specialized time series prediction model based on Mamba, designed for accurate cluster workload forecasting, surpassing current benchmarks like the Transformer model in specific tasks. Additionally, we present the RL-based Power Control Policy (RPCP), which autonomously adjusts server power states based on predictive load information, effectively reducing energy consumption while meeting latency requirements. Our study evaluates various VM resource allocation algorithms, identifying approaches that optimize power consumption and latency tradeoffs under ideal conditions. Experimental results on real workload traces show that our proposed Attentive-Mamba predictor and RPCP significantly improve the energy efficiency of the data center and maintains the required cloud system performance compared to some other VM consolidation algorithms.