Research On Auto-Scaling Strategies For Heterogeneous Resources In Cloud Environments

Currently,cloud computing has become a widely used paradigm for deploying web-based applications.More and more public users tend to deploy and manage their web-based services on public cloud platforms.The most appealing sell point in cloud computing is elasticity.With the help of elasticity,cloud application providers can dynamically adjust the type and quantity of virtual machine(VM)instances provided by the cloud resource providers to optimize the renting cost while meeting the users’ quality of service(QoS).The process of automatically scaling VMs on the cloud environment is referred to as auto-scaling.It is of vital importance for cloud application providers to obtain optimal auto-scaling policies when running web-based applications in the cloud environment.However,the diversity of cloud resources and the uncertain nature of user workloads in dynamic cloud environments make the heterogeneous resource auto-scaling problem more challenging.Based on the above,this paper mainly focuses on the two directions of user workload forecasting and adaptive heterogeneous resource scaling.The main contributions are summarized as follows:To the best of our knowledge,the first time use the Transformer network based on the multi-head attention mechanism to predict user workloads in dynamic cloud environments.With the help of this work,cloud application providers can obtain a more scientific heterogeneous resource scaling policy based on the predicted workloads,thereby meeting QoS while reducing instance rental costs.Based on the Wikimedia workload dataset,this work conducts a large number of experiments to compare the performance between our prediction algorithm and other benchmark algorithms using root mean square error(RMSE)and average absolute percentage error(MAPE)as evaluation metrics.The results show that the accuracy of our proposed prediction algorithm is better than benchmark algorithms.Based on the advantages of reactive and proactive algorithms,and combining the flexibility of on-demand VMs and the low price of reserved VMs,this thesis proposes a hybrid deep reinforcement learning-based auto-scaling algorithms for helping cloud application providers adaptively obtain the heterogeneous horizontal scaling policy without prior knowledge.With the help of HDRLAS,cloud application providers are able to guarantee public users’ QoS while minimizing instance rental costs in cloud environments with uncertain user workloads.Experiments show that our HDRLAS can significantly improve resource utilization and achieve better cost-efficiency than other auto-scaling algorithms.