Research On Scheduling Algorithm Of Heterogeneous Multi-core Computing Computing Cluster For NFV

5G,which has been promoted in recent years,continues and develops the modification of the previous generation of mobile communication systems on the Radio Access Network,which uses Network Function Virtualization technology to virtualize the RAN,thereby reducing the network upgrading complexity and capital investment in specialized communication equipment.Since the computing architecture of the virtualized computing platforms,such as cloud computing,is usually heterogeneous.Therefore,designing a task scheduling algorithm for vRAN(virtualized RAN)task flow that can be applied to heterogeneous computing platforms will have an important impact on the performance of the vRAN.In addition,computing platforms often bring huge energy consumption when running.In order to meet the carbon neutrality concept,when designing task scheduling algorithms,energy consumption needs to be taken as a key algorithm performance indicator.This paper mainly studies the task scheduling problem of the physical layer,and proposes a three-stage scheduling algorithm for energy optimization and a task graph generation algorithm for vRAN.The main work and innovations of this paper are as following:1.By analyzing the layer slicing option of the physical layer in the vRAN environment,a task graph generation algorithm with controllable randomness is proposed,which can reflect the characteristics of the actual PHY signal processing flow.This algorithm is based on the existing random task graph generation algorithm.By analyzing existing algorithm designs,the key requirements that must be met when designing the task graph generation algorithm are obtained,which are,the generated task graph structure and the weights of the nodes and edges of the graph must be random.Since the randomness of the task graph obtained by the existing scheme is uncontrollable,this paper designs and implements the task graph generator by itself.2.By analyzing the logic of the existing classic list scheduling algorithm HEFT(Heterogeneous Earliest Finish Time)with time as the main performance indicator,it is transformed to make the scheduling results under the premise that it has a slight impact on the total execution time,but can give more idle time slots for use in the energy consumption optimization stage.Then,by using the local idle time slots in the initial scheduling results,an attempt is made to find the core with the optimal energy consumption and re-schedule the task on that,and the core with optimal energy consumption is able to reducing more energy consumption while extending the same amount of execution time.Finally,this paper uses the Dynamic Voltage Frequency Scaling technology to try to decrease the frequency of the core which results in prolonging the task execution time but reducing more energy consumption,and the DVFS technology will be applied on the remaining idle time slots.The combination of the above three steps is the proposed three-stage scheduling algorithm.Simulations show that the proposed algorithm can obtain nearly optimal scheduling results in energy consumption under the condition of meeting the given deadline.