Research On Delay Aware Game Operation Cost Optimization Strategy In IaaS Cloud Environment

With the rapid development and maturity of the IaaS platform,some small and medium-sized game operators began to deploy game applications by renting virtual machine instances from IaaS providers.Because the virtual machines in the IaaS platform share the hardware and software resources of the data center,serious competition for shared resources and performance interference problems will occur between the virtual machines,which in turn will cause the performance of the game applications deployed on it to significantly decrease,greatly affecting Player's gaming experience.In addition to performance issues,game operators also need to pay attention to the rental cost of virtual machines.If the resources of the virtual machines leased far exceed the actual needs,the resources cannot be fully utilized,which will lead to unnecessary cost increases;If the leased virtual machine resources are less than the demand,the player's game experience requirements cannot be guaranteed.In view of this,we study the latency-aware game operation cost optimization problem in the IaaS environment.By combining online player request allocation and dynamic virtual machine provisioning mechanism,the virtual machine rental cost is minimized while ensuring game response time.Specifically,this paper completed the following works:(1)We designed and implemented an online player request allocation algorithm based on LSTM,which uses the LSTM neural network model to predict the end time of player sessions,and assigns players with similar session end times to the same virtual machine in order to reduce The lease time of the virtual machine thus reduces the virtual machine rental cost of the game operator.(2)We designed and implemented a dynamic virtual machine supply algorithm based on queue theory.This algorithm uses the queue network model to predict the response time of the game,and dynamically adjusts the number of leased virtual machines according to the predicted response time,in order to ensure in the case of game response time,the game operator's virtual machine rental costs are minimized.(3)Under the CloudSim platform,simulation experiments were conducted using World of Warcraft,World Heritage Warfare and Tank World datasets.Experimental results show that compared with other existing mechanisms,the proposed mechanism can save up to 35% of the cost,while also providing a sufficiently good response time.