The Design And Implementation Of The Aggregation And Analysis System Of SDN-Campus Controller's Performance Data

The current SDN-Campus controller has been widely used in traditional park,chain retail,chain hotel,chain restaurant and other centralized management scenes.However,with the development of the SDN market,the controller needs more and more network devices,therefore,the original scenario design scheme with small capacity cannot carry the massive performance data of the monitor network devices.In this paper,it mainly focus on how to manage storage aggregation performance data efficiently and do statistics and prediction analysis by introducing the aggregation and analysis design scheme which is redesigned by big data technique.The following problems are solved in the aggregation and analysis system of SDN-Campus controller's performance data: first,by comparing different mass data storage systems,HBase is selected as the database for storing mass performance data.According to the time dimension characteristics and space dimension characristics of performance data,it optimize the design of Rowkey,and solve the query problem of mass performance data by improve the storage and query efficiency of performance data through custom Filter and BulkLoad.Secondly,by comparing different computing frameworks and using HBase as a comprehensive consideration of mass data storage systems,Spark is selected as the mass data computing platform for aggregating and analyzing the mass performance data.By using RDD and DataFrame structure,MapReduce operation,Spark-SQL technology,and Java and Scala mixed programming model,the system completes the complex computing tasks of mass performance data,and further improves the feasibility and efficiency of the calculation process.Finally,in connection with the characteristics of the performance data in a restaurant scene,the system constructs the Long Short-Term Memory network LSTM model to predict the performance data,and send the tasks of TensorFlow to Spark with the TensorFlowOnSpark component.In this way,the prediction time of the prediction model is largely reduced.At present,the development of the auxiliary function module and aggregation analysis core function module has achieved in the aggregation and analysis system of SDN-Campus controller's performance data.After the test and verification of the project team,the system has met the expected requirements.The implementation of the aggregation and analysis system of SDN-Campus controller's performance data provides a better solution for other systems of SDN-Campus controller to storage and calculate the mass data.It also provides a reference for the project team to upgrade the small capacity design to large capacity design in the future.