Enhancing virtual machine placement efficiency in cloud data centers through fluctuations-aware resource management

The optimal placement of virtual machines in data centers holds significant importance. Failing to address this matter accurately may lead to an increased number of failures and frequent migrations between physical machines to accommodate the new resource requirements of virtual machines based on the evolving workload. This research focuses on predicting future resource fluctuations. Therefore, in our proposed model, virtual machines are categorized as either 'requiring additional resources in the future' or 'requiring fewer resources or no change in the future.' Consequently, virtual machines with varying labels, referred to as complementary, are placed accordingly. The primary objective of this study is to predict and monitor the service requirements of an organization's users. To achieve this goal, time series data and LSTM and GRU algorithms were employed. These algorithms were applied to multiple datasets to train a resource prediction model and subsequently utilize it for categorizing new requests. The results demonstrate that the proposed model has reduced the number of migrations by a maximum of 31% compared to the Best Fit Algorithm and a maximum of 25% compared to the Worst Fit Algorithm for 32,500 requests, encompassing both initial placements and changes in resources after the initial placement. In addition to its predictive capabilities, the proposed model contributes to enhanced resource allocation efficiency, ensuring optimal usage of data center resources. By leveraging advanced machine learning techniques, the model demonstrates its effectiveness in accurately anticipating future resource requirements and minimizing the overall operational overhead, as well as reducing placement failure by 2% compared to the Best Fit algorithm.