Research And Implementation Of Network Resource Scheduling Architecture In Hybrid Data Center

The data center is the core infrastructure supporting technologies such as cloud computing and big data.It provides computing resources,network resources,storage resources and other support for these technologies to ensure users’ on-demand access anytime,anywhere.With the increase of user demand,the scale of the data center is also expanding,and it also faces the problem of low resource utilization.One of the main reasons for the low resource utilization is the low efficiency of resource scheduling.The currently widely used resource scheduling methods have the following problems: First,the resources are scheduled according to the nominal value,in order to maximize the service quality of the business,most of the existing scheduling architectures plan and design resource allocation schemes based on the nominal value of business requirements,although this resource allocation method can ensure the stable operation of the business,it causes a lot of resource redundancy.The reason is that most businesses usually only reach the nominal value in a few cases;second,the static scheduling of resources.Most of the existing resource allocation schemes are static and immutable.Even if the actual load of the business is much lower than its nominal demand,the resource allocation result cannot be dynamically adjusted;third,static networks are difficult to adapt to dynamic services.In view of this,this paper studies and implements the network resource scheduling architecture of the hybrid data center.The main work and innovative achievements are as follows:For the traditional data center network,the problem of low resource utilization caused by the incompatibility of static and conservative resource allocation methods with time-varying services is analyzed;the existing business load forecasting and resource scheduling technologies and their applicable scenarios,advantages and disadvantages are reviewed;at the same time,it focuses on the evolution process of data center network architecture,virtualization technology and typical time series prediction methods,which provides theoretical support for designing a closed-loop scheduling architecture with high resource utilization.Aiming at the problem of low utilization of data center network resources,a closed-loop scheduling architecture of hybrid data center network resources with high resource utilization is proposed.Firstly,the necessity and feasibility of the architecture are analyzed,and a closed-loop scheduling architecture model is established.The architecture mainly includes the following modules: 1)The sensing module is responsible for the collection of business load information,which is realized through the probe program;2)the forecasting module is responsible for business load forecasting.An adaptive Kalman filter algorithm is designed to forecast business load,and a model that conforms to business characteristics is established.When the forecast deviation is large,the predicted value is "initialized" to reduce the forecast deviation in the future,to improve the adaptability and prediction accuracy of the algorithm;3)the reconfiguration module is responsible for the redistribution of business resources.By using the predicted value of resource utilization and the predicted value of business load output by the prediction algorithm,a reconfiguration strategy based on usage is constructed to improve resource utilization.The simulation results based on the Alibaba data set show that the adaptive Kalman filter algorithm proposed in the prediction module has higher adaptability and flexibility than the basic Kalman filter algorithm,the prediction accuracy is more accurate,and the model prediction accuracy is improved by 42%;after using the proposed prediction method,the overall resource utilization rate after reconstruction is increased to more than 80%,which verifies the effectiveness of the proposed closed-loop scheduling architecture.For the problem of how to realize the closed-loop scheduling architecture with high resource utilization,the intelligent wired and wireless hybrid scheduling software was designed and developed,and the interface was designed to integrate with the container orchestration system Kubernetes,and a closed-loop based hybrid data center network resource scheduling platform was constructed.Firstly,by analyzing the design requirements of the platform,its framework is established,and the Kubernetes cluster environment of the platform is introduced;Then,the software and its functional modules are introduced in detail.The software mainly realizes the perception function,prediction function and business reconstruction function of business load,and integrates with Kubernetes to realize resource scheduling to complete the closed-loop design of "perception-prediction-reconstruction".Finally,a platform and a test scenario were built in a company’s experimental bed and data center,and the feasibility and efficiency of the hybrid data center network resource scheduling architecture were verified by testing with real data.