Research On Key Optimization Technologies Of Task Scheduling And Computing Resource Allocation In Cloud-Fog Computing System

With the development of the Internet of things and the mobile Internet,the cloud-fog computing has gradually replaced the traditional cloud computing as the new important infrastructure of the Internet.Its quality of service(Qo S)and resource usage efficiency have also become the major concerns of the network services.The task scheduling and computing resource allocation strategy of the cloud-fog computing system is the main factor which affects its quality of service and resource usage efficiency.Thus,it is necessary to optimize the task scheduling and computing resource allocation in the cloud-fog computing system.Because of the distributed architecture of the cloud-fog computing system and differences between the fog node and cloud center in system architecture,computing resources and network transmission,the service of quality of tasks and resource usage efficiency of the cloud-fog computing system optimization not only needs to consider its overall task processing strategy,but also needs to consider the task scheduling and computing resource allocation strategies of the fog node and the cloud center respectively.Therefore,in order to optimize the quality of service of tasks and resource usage efficiency of the cloud-fog computing system,this paper focuses on three aspects to research: workload allocation and task scheduling of the cloud-fog computing system,task scheduling and computing resource allocation of the fog node as well as the cloud computing resources on-demand allocation and elastic scheduling,which includes:1.To improve the service of quality of tasks and resource usage efficiency of the cloudfog computing system,this paper proposes a delay-aware online workload allocation and task scheduling algorithm based on distributed architecture by using Lyapunov drift-plus-penalty optimization method.Specifically,the proposed algorithm can reduce the task service delay of the cloud-fog computing system through optimizing the overall task processing strategy of the cloud-fog computing system.According to the workload of the fog node and the real-time state of the network,the proposed algorithm can obtain the optimal workload allocation and task scheduling strategy among different devices in the cloud-fog computing system by weighing the queuing delay in the fog node and the task network delay,so that different devices in the cloud-fog computing system can cooperate with each other,which can reduce the task service delay on the premise of making full use of the computing resources of the cloud-fog computing system.Moreover,the proposed algorithm enables each fog node to estimate the workload state of its neighbor fog nodes through the single information broadcast in the previous time slot,which avoids the traversal operation and frequent information interaction of all fog nodes in the system under the centralized management mode.Thus,the proposed algorithm realizes the loose coupling of the information among different fog nodes in the distributed architecture to achieve that each fog node can manage its workload allocation independently,which satisfies the deployment requirement of the cloud-fog computing system distributed architecture.2.To improve the service of quality and resource usage efficiency of real-time heterogeneous tasks in the fog node,this paper proposes the corresponding algorithms for optimizing the task scheduling and computing resource allocation of the fog node.Firstly,according to the limitation of computing resource capacity of the fog node,this paper proposes an adaptive queue weight-based computing resource allocation of the fog node by using Lyapunov optimization method.According to the length of different type task queues,the proposed computing resource allocation algorithm can allocate the computing resources of the fog node to the tasks in the longer queue first,which not only avoids the problem of task “resource starvation” caused by the imbalance of computing resources allocation in the fog node,but also guarantees the fog node throughput optimization.On the basis of analyzing the impact of task sorting in the task queues on the task processing performance of the fog node,this paper proposes two task buffering sorting scheduling algorithms of the fog node with the task execution time evaluation upper bound and laxity time as the standard,i.e.,the multiobjective and non-linear multi-objective monotonically increasing task buffering sorting scheduling algorithms of the fog node.The two proposed task buffering sorting scheduling algorithms can optimize the task scheduling in the task queue of the fog node to achieve the ondemand optimization tradeoff between the fog node throughput and the system task completion ratio according to different performance requirements on the fog node throughput and system task completion ratio.3.To improve the computing resource usage efficiency of the cloud center and avoid the computing resource allocation and scheduling lag behind the workload change,this paper proposes a cloud computing resource on-demand allocation and elastic scheduling method based on proactive mode.Firstly,this work uses the task request number workload(i.e.,network workload)of Wikimedia service as the research objective and proposes an adaptive two-stage multi-neural network workload prediction method based on long short-term memory(LSTM)model.The proposed workload prediction method can classify the input network workload data into the uphill and downhill categories according to the input network workload data trend characteristics.Then,according to the classification result,the input network workload data will be adaptively scheduled to the corresponding LSTM workload prediction model to obtain the network workload prediction result at the next time point,which can improve the network workload prediction accuracy.Furthermore,this paper proposes a maximum cloud service profit resource search algorithm based on network workload prediction.Considering the quality of service of tasks and the stability of the system,the proposed algorithm can obtain the required cloud server number with improving the cloud service profit as the optimization objective according to the workload prediction result.It not only makes the cloud computing resources be allocated and scheduled in advance to avoid the impact of computing resource allocation and scheduling lag behind the workload change on the service of quality of tasks and resource usage efficiency of cloud computing,but also realizes the cloud computing resource on-demand allocation and elastic scheduling with the goal of improving the cloud service profit.