A Comprehensive Review Of Machine Learning Approaches For Dynamic Resource Allocation In Multi-Tenant Cloud Environments

Abstract

Its static nature has made it challenging to provide elastic resource allocation in multi-tenant cloud systems where workloads evolve and tenants have different requirements. This paper introduces a new scheme of dynamic resource allocation based on a combination of machine learning methods that allows for achieving resource economy and improving the general cloud performance. The research in this work is concerned with creating predictive models which exploit historical workload information, tenant usage behavior and application characteristics for accurate reconciliation of resource requests. Through the analysis of these factors, the proposed system adaptively assigns the resources (e.g., VM, CPU, memory and storage) automatically for the tenants’ demands that are changing over time, with better performance and efficient resources. The study also investigates the application of reinforcement learning algorithms in dynamically or proactively optimising the resource allocation policies in response to online feedback and environment changes coming from real use case scenario. The strategy is characterized by scaling out and adapting to different workloads and tenant preferences, by learning from the interactions of the system and through fine-tuning resource allocation decisions. Experimental results on a multi-tenant cloud testbed show that the machine learning based resource advancement system is efficient. It is found that the resource utilization, the response time, and the overall system performance are much better with dynamic resource allocations than with the traditional static approach. In general, our research aids the progress of cloud computing by offering a proactive and intelligent way for the resources allocation in multi-tenant setting. The system, implemented as a comprehensive set of machine learning algorithms, allows cloud providers to effectively monitor their resources, extract patterns and dependencies, optimize performance provisioning and address varied tenant requirements, while eventually improving the overall quality of service of the cloud.