Research On End-to-end Flow For Time Sensitive Network

Time Sensitive Network(TSN),as a set of IEEE standards,extends Ethernet to meet the strict time constraints of modern critical task applications by providing real-time and low-jitter communication capabilities.Due to its real-time and high bandwidth communication capabilities,TSN is a promising communication technology in many application fields,such as industrial automation and vehicular networks,due to its realtime and low-jitter communication capabilities.The core mechanism of TSN is flow scheduling,and the efficiency of flow scheduling algorithms is an important consideration factor.Heuristic algorithms are generally used to reduce computational complexity and improve solving speed in existing TSN flow scheduling.However,research has shown that input order of data flows is also a key factor affecting the solving time of flow scheduling algorithms.This study only verifies the impact of input order on flow scheduling algorithm solving time,but in reality,the influence of flow input order on the solving time of flow scheduling algorithm is complex,which making it challenging to obtain a good order to reduce the solving time of flow scheduling algorithms.Therefore,in order to reduce the scheduling time of TT flow and the total runtime of TT flow scheduling(including TT flow sorting time and flow scheduling time).this paper focuses on the time-triggered(TT)flow input order perspective to search for methods to improve the efficiency of TT flow scheduling.The main work of this paper includes:(1)A genetic-tabu hybrid search algorithm is utilized to explore the input order of TT flow and reduce the runtime of flow scheduling algorithms.Based on the SMT flow scheduling algorithm,this genetic-tabu hybrid search algorithm is used to determine the input order of TT flow when scheduling.The algorithm uses the better global optimization ability of the genetic algorithm to generate the initial solution of the TT flow input order,and constantly generates new individuals that adapt to the environment through selection,crossover and mutation operations,so as to obtain a satisfactory solution.The taboo operation is used to improve the efficiency of the search by controlling the crossover and mutation based on the superior neighborhood search ability of the tabu search algorithm.(2)A deep reinforcement learning framework(PSNDRL)for quickly sorting TT flows is proposed to reduce the total running time of TT flow scheduling,including TT flow sorting time and flow scheduling solution time.This paper designs a deep reinforcement learning framework for quickly sorting TT flows,trains and uses a sorting agent network to sort TT flows.This framework includes three key modules: the preprocessing module for creating the relationship graph between TT flows,the agent module for mining and quantifying the complex correlation between TT flows and selecting the TT flow with the highest probability value,and the environment module for scheduling and reward calculation of TT flows.(3)Based on the in-vehicle network dataset,the experiment shows that the input sequence of TT flows found by the hybrid genetic-tabu search algorithm is superior to the traditional genetic algorithm and Tabu search algorithm in terms of scheduling time optimization,compared with the genetic algorithm and Tabu search algorithm,the flow scheduling time of the hybrid genetic-tabu search algorithm is reduced by 7.74% and10.74%,respectively.The flow total scheduling time(including flow sorting time and flow scheduling solution time)of PSNDRL is reduced by 25.95% and 24.62% on average,respectively,compared with direct scheduling without sorting and methods experience of sorting by flow attributes,and compared with the hybrid genetic-tabu search algorithm,the flow scheduling time is also reduced.