Research And Application Of Workload Forecasting And Task Offloading For Cloud-edge Computing Systems

With the development of information society and the rapid improvement of the Internet of Things,the use of network data is growing explosively.A new computing model named Cloud-Edge Computing is formed by Cloud Computing and Edge Computing.However,due to the growing demand for computing,the difficulties faced by Cloud-Edge Computing are also increasingly highlighted.On the one hand,data centers have more complex architecture designs and provide more diversified computing services than before.It is of great significance to establish an appropriate workload forecasting model based on the collectible data features,which is also the prerequisite for resource scheduling in data centers.On the other hand,since the computing requirements are unpredictable,the traditional task offloading strategies may bring network congestion frequently,which leads to the number of failed tasks increasing,the task execution time prolonging and Qo S quality decreasing.Designing an effective task offloading method will affect the processing capability directly,which is also helpful for improving system performance and meaningful for research value.To solve the above two problems,this thesis focuses on workload forecasting and task offloading and studies the implementation of load forecasting and the decision-making scheme of task offloading in Cloud-Edge computing.The main work includes the following aspects:(1)Most of the current load forecasting algorithms are designed only for a single data center,and cannot balance the dynamic dependence between the long-term trends and the local characteristics.So,we propose a load forecasting model for the multi-data center.The model considers the dynamic dependence between input features and load forecasting results,as well as the long-term trend of load fluctuation.Moreover,the autoregressive model is used to model the linear trend of load series,and the neural network model is used to capture the nonlinear trend of time series.(2)Researchers ignore the collaboration among multiple resources and only optimize the performance of the system with a single objective like time delay or energy consumption.So,we propose an adaptive fuzzy logic-based collaborative task offloading scheme.Firstly,the multi-dimensional collaboration between servers,cache services,and cloud or edge devices is designed.Secondly,the method includes an adaptive fuzzy logic-based offloading algorithm,which can revise the offloading decision constantly.Experimental results show that the proposed scheme can provide reliable performance while dealing with task offloading under the MEC environment.While compared with other state-of-the-art,the performance is improved in reducing the failure rate of task and processing time.(3)The above algorithms are implemented and the computing monitor system for cloud-edge computing is constructed.The system integrates the proposed load forecasting model for the cloud multi-data center.It supports selecting different methods to obtain the load forecasting result.The load change trend and final result are visualized to the user.The system also implements the adaptive fuzzy logic-based collaborative task offloading scheme.We can send a task offloading request on the visual interface,and query relevant offloading information,which includes the clustering result of edge servers,orchestrator scheduling information,computing results in cache,and the fuzzy rules used in the process.