محمد يكتا

With the advancement of cloud computing technology, there is an increasing demand for maximum utilization of cloud resources, which has led to significant energy consumption in cloud data centers. Resource provisioning is one of the challenging problems in the cloud environment. Resources should be dynamically allocated according to changes in demand in the application. Over-provisioning wastes energy and costs. On the other hand, under-provisioning causes violation of service level agreements (SLAs) and degradation of Quality of Service (QoS). Virtual machine consolidation is part of resource provisioning in cloud computing. Virtual machine consolidation provides an approach to reduce the energy consumption of cloud data centers. Consolidation of virtual machines can be done more effectively by predicting the future workload of cloud datacenters. The process of virtual machines consolidation consists of four steps, including the following: (1) detection of overloaded hosts, (2) detection of underloaded hosts, (3) virtual machines selection for migration, and (4) allocation of virtual machines. We have provided a method for the two steps of detecting overloaded hosts and choosing a virtual machine for migration. The purpose of these algorithms is to reduce energy consumption and SLA violations and reduce the number of migrations of virtual machines. In the first part of our work, which is the detection of overloaded hosts, we first used three models: LSTM, BiLSTM, and GRU, to predict the amount of CPU usage in virtual machines and compared these three models using the MSE, MAE, and Huber Loss eva‎luation criteria. Finally, we used the obtained results to detect overloaded hosts. In the second part of our work, choosing a virtual machine for migration, we presented an optimal energy algorithm. This algorithm considers the CPU and RAM usage of each virtual machine and selects the virtual machine with the highest score for migration. The energy consumption, number of virtual machines migrations, and the combined eva‎luation criterion of ESM were improved compared to other algorithms. In the last part of our work, we combined overloaded host detection and the virtual machine selection algorithm. Compared to LR-MMT algorithm, we saw a reduction in energy consumption, and the number of migrations of virtual machines, SLAs, and ESM by 24%, 86.1%, 61.8%, and 96%, respectively.

رايانش ابري , يكپارچه سازي ماشين هاي مجازي , پيش بيني بار كاري , يادگيري عميق

cloud computing , virtual machine consolidation , workload prediction , deep learning

Mohammad Yekta

Dr. Hadi Shahriyar Shah hosseini