مريم حيدري

Cloud computing provides on-demand access to powerful computing resources, storage, or bandwidth over the Internet. There are two participants in cloud-based systems: service providers, who are resource owners, and service users or tenants. Users dynamically enter and exit the cloud environment at different times to request hosted resources. Service providers allocate requested resources to incoming users. This resource provisioning process is an important part of cloud data centers. The variation observed in the number of users accessing cloud services at different times makes provisioning a challenging problem for service providers. Inefficiency in resource allocation leads to over/under provision of resources. Insights into future resource utilization trends serve as an important aid to service managers. Cloud resource workloads are highly heterogeneous and dynamically varying in nature. Adaptive models play an important role in modeling real-time processes that exhibit timevarying behavior. The computing load on a data center is dynamically driven by the varying demands of users. Training one-time workload prediction models using past trends in workload changes is not enough to model future workload change trends. Predictive models operating in such environments must adapt to new patterns, otherwise their accuracy will decrease. Incorporating new recorded data into built models can improve model prediction accuracy and can generate better resource utilization predictions. Also, there are different features in the collected data, and the selection of the most important features has a great impact on increasing the accuracy of forecasting the use of resources. Therefore, in this research, a model based on CNN-BiLSTM has been used to predict the use of resources. Also, two well-known Pearson and Spearman correlation methods have been used to select the best features. For eva‎luation, the proposed method has been implemented on the Google cluster dataset, and the results were eva‎luated with three eva‎luation criteria: mean square error, mean absolute error, and R2. The error of 0.001 in the eva‎luation results shows that the proposed method has worked very well to predict the consumption of cloud resources.

پيش بيني مصرف منابع , محاسبات ابري , حافظه كوتاه مدت بلند مدت , ضريب همبستگي پيرسون , ضريب همبستگي اسپيرمن

resource consumption prediction , Cloud computing , Bilstm , Pearson correlation coefficient , Spearman correlation coefficient

Maryam Heidari

Mohsen Sharifi