Research On Hybrid Cloud Resource Allocation Algorithm Based On Markov Decision Process

Users have increasingly demanded computing resources in recent years with the rapid development of information technology.Traditional computers have restricted modern office and applications to a certain extent in terms of computing capacity,storage capacity,and convenience.Cloud computing provides users with unlimited computing and storage capabilities,users can use it anytime,anywhere through the network,more flexible to meet user needs.Enterprises or research institutes with private clouds can use public clouds to flexibly expand computing resources to cope with instantaneous request overflows and unbalanced loads.Therefore,hybrid cloud has become the first choice for enterprises or research institutions to deploy cloud computing applications.Therefore,how to allocate resources in the hybrid cloud reasonably has been a hot issue in recent years.There are still some problems in the existing research: Does not take into account that the arrival and departure of task requests in the hybrid cloud platform are dynamically changing.The different types of task requests have different resource requirements for Virtual Machines(VM).The execution time of the task is also uncertain.In the private cloud,the utilization rate of VMs is low,and resource waste is serious.The revenue of users in the hybrid cloud are not considered from a long-term perspective,and it is easy to fall into a local optimal solution,resulting in lower revenue for users in the hybrid cloud.This thesis mainly focuses on the resource allocation problem of the hybrid cloud platform.By designing appropriate models and algorithms,the resources in the hybrid cloud are reasonably and effectively allocated to improve the utilization rate of VM in the private cloud and maximize the users' revenue.Suppose there are two different task requests in a hybrid cloud,their arrival and departure are dynamic.The execution time of the task is also uncertain.The number of VM in the private cloud changes with time.Markov Decision Process(MDP)theory and Value Iteration Algorithm(VIA)are used to solve above problems.The main contributions of this thesis are as follows:(1)Proposed a hybrid cloud resource allocation model and algorithm based on MDP.Using the MDP theory,the hybrid cloud platform resource allocation problem is modeled.First,establish the state space,action space and revenue model of the hybrid cloud platform.Second,derive the state transition probability of the system and establish a hybrid cloud discount revenue model.Finally,according to the Bellman equation,derive the user's expected revenue in the hybrid cloud equation.Use VIA to solve the user's expected revenue in each state,select the maximum expected revenue,and get the optimal strategy.Experimental results show that the hybrid cloud resource allocation algorithm based on the MDP model proposed in this thesis can enable users to obtain the maximum revenue in the hybrid cloud.(2)Proposed a hybrid cloud resource allocation model and algorithm for public cloud prices with time.In order to better fit the actual situation,a new public cloud pricing method is proposed.The price of public cloud changes with time.Design the price formula and time step of public cloud,and improve the model and algorithm of resource allocation of hybrid cloud platform.Through comparative experiments,Experimental results show that after the price improvement,the model can still maximize the revenue of the user in the hybrid cloud.And analyze the hybrid cloud strategy from the aspects of ordinary and confidential task request arrival rate,number of VMs in private cloud and VM release rate.The analysis proves that the MDP model proposed in this thesis can dynamically allocate resources according to the state of resources in the hybrid cloud to maximize user revenue.