Research Of Adaptive Resource Allocation In Cloud Gaming

Recently,cloud gaming has become increasingly popular.The interaction delay of cloud gaming is closely related to players' experience of the game.Thus,how to reduce the interaction delay has become one of the biggest challenges of cloud gaming.The interaction delay caused by the response on the server side is greatly affected by how the process allocates the resources.However,it is difficult to find the optimal resource allocation strategy that minimizes response delay on the server side.In this paper,we propose an adaptive resource allocation method that effectively and adaptively allocates resources online to reduce response delay on the server side.Firstly,through machine learning,we build a performance model which can capture the complex relationship between resource partition and system performance.Using this model,we divide the processes into disjoint groups and partition resources among process groups,which greatly simplifies the problem of resource allocation while making it efficient.Then,in order to tackle dynamic workload changes,we quantify and cluster system workload,use reinforcement learning to learn how different resource partitioning actions affect system performance during the online phase,and adaptively select the best action to minimize response delay in real time.We use several real games to evaluate our method in a real cloud gaming environment.The experimental results show that the adaptive resource allocation method can reduce the response delay by 20%-41% compared with the system without resource partitioning,and is significantly better than other resource allocation methods.