Virtual Machine Scheduling And Reservation Strategy For Downlink Joint Transmission Of CoMP In C-RAN

In centralized baseband pool of C-RAN architecture, a number of baseband processing units are centralized. Multipul computing resources such as CPU, memory and bandwidth are encapsulated by adopting the virtualization technology and then provide service for users in form of virtual machines(VMs). Users schedule their computing tasks on corresponding VMs to run and pay to cloud service providers. In order to improve the system spectrum efficiency, reduce users' cost and guarantee service fairness among users, we should study efficient virtual machine scheduling and reservation strategies for JT CoMP in C-RAN. In this paper, the main work and innovation are as follows:For JT CoMP scenario, an efficient VM scheduling strategy in C-RAN is proposed. In CRAN baseband pool, each user's computing tasks are composed of a number of interrelated protocol function modules. A module must be mapped to corresponding type of VM in accordance with its task nature. In order to realize joint synchronization of all cooperating RRUs and service fairness among users, we define the objective function for minimizing the maximum completion time among users, while taking a full consideration of the associativity among modules, the delay constraints, the resource constraints and mapping constraints.To solve the problem, here we consider the three static scheduling algorithms: Genetic Algorithm(GA), Genetic Simulated Annealing Algorithm(GSAA) and Pacing Algorithm(PA).Finally, the effectiveness of the proposed strategy is demonstrated by experimental simulation.With less number of users, single business and small amount of calculation, a VM reservation strategy for a single cloud service provider is proposed. For simplicity, we assume that the probability of the number of user arrivals is consistent with Gaussian distribution. Then,we calculate the best combination of the three types of VMs that will incur the smallest cost by means of probability weighted according to the used numbers by GSAA as well as the given pricing model. Finally, the effectiveness of the proposed strategy is demonstrated by experimental simulation.With more number of users, multipul business types and large amount of calculation, a VM reservation strategy for several cloud service providers is proposed. At first, we predict the number of users and the resource demands of users' modules for future time slots according to the data in a time period in the past. Then the total numbers of demand for each type of VM are toughly calculated and continually adjusted according to users' delay constraints. Next, the total numbers of VMs are allocated to different service providers while satisfying the resource constraint of each service provider. And then, we calculate the best combination of the three types of VMs for each service provider that will incur the smallest total cost.Finally, the effectiveness of the proposed strategy is demonstrated.