Researches On Technologies Of Dynamic Routing And Spectrum Assignment

With the rapid development and application of bandwidth-thirsty data services,especially cloud-based services,network traffic shows an explosive growth and dynamic change trend.Elastic optical networks(EONs)with high spectral efficiency and flexible spectrum resource allocation have emerged.The routing,modulation level and spectrum assignment(RMLSA)technology in EONs can arrange the best RMLSA scheme for incoming services according to current information of network resources and services,which can achieve effieicnt utilization of network spectrum resource while meeting the service quality requirements.Therefore,RMLSA has become one of the research hotspots in EONs.In this dissertation,the research mainly focuses on the RMLSA in EONs,including the planning and provisioning of services with different communication schemes,the comprehensive optimization of the overall network performance and the guarantee of network survivability.The main work and innovative contributions are listed as follows.1?In EONs,manycast as a flexible communication scheme of point to multipoint providing efficient communication guarantee for data backup,cloud computing and etc.Most of the existing manycast RMLSA algorithms are not suitable for manycast services,due to they are directly improved based on multicast RMLSA schemes,resulting in lower spectrum utilization and higher network energy consumption.To solve the above problems,this dissertation proposes a manycast RMLSA algorithm based on genetic algorithm framework(MA-RMLSA-GA)to jointly optimize the destination node selection and spectrum resource allocation to achieve efficient spectrum resource utilization.Moreover,a manycast RMLSA algorithm based on unicast request decomposition coding(URD-GA)is proposed to reduce the energy consumption of manycast tree by searching for links that are already in working state to form a routing optical tree,so that as many links as possible in the network are dormant.The simulation results show that MA-RMLSA-GA algorithm reduces the request blocking probability by up to 84%,and URD-GA algorithm makes the network energy saves up to 23%,compared with the reference algorithms.2?Various new technologies and services in EONs continue to appear.While the network carries more traffic,the energy consumption also increases rapidly.As two important indicators to measure network performance,network service capacity and network energy consumption are optimized separately in the existing RMLSA algorithm.To address this problem,this dissertation first defines the energy efficiency value to evaluate the network performance qualitatively.Then,an independent optimization genetic algorithm strategy(IOGA)and a priority-based scheme rationality adjustment strategy(PRA)based RMLSA algorithm(IOGA-PRA)is proposed to optimize energy efficiency in hybrid-cast scenarios.The IOGA strategy is used to search for the optimal RMLSA scheme for each hybrid-cast request in the current network state independently,and the PRA strategy is used to eliminate the spectrum conflict between the independent RMLSA schemes for the batch of requests.The simulation results show that the IOGA-PRA algorithm improves the energy efficiency by up to 86%in hybrid-cast scenarios,compared with the reference algorithms.3?In the process of rapid development of inter datacenter EON,the network operators pay more attention to the dimension of network performance,comprehensive evaluation and optimization of the overall network performance,not limited to the improvement of certain network performance indicators.It is one of the keys to cope with the transformation and upgrading of business and technology in EONs.Furthermore,the existing RMLSA algorithms of inter datacenter don't utilize the characteristics of content connectivity services in the network to optimize the network performance supporting survivability,which results in the decline of the overall network performance.In this dissertation,an evaluation index(i.e.,WCES)is defiend to measures the overall network performance in network transmission capacity,the number of served customer and survivable network cost.Furthermore,an anycast RMLSA algorithm based on contentization of network node and topology(CNNT_RMLSA)is proposed.By using the changeable characteristics of destination node of the content connectivity service,the actual physical topology is simplified to the virtualized network topology after the contentization,and the two-way selection between the datacenter and the user is realized while improving the survivability of network.The simulation results show that the validity of the WCES index and the CNNT_RMLSA algorithm can effectively improve overall performance while ensuring network survivability.The simulation results show that the WCES index is effective and CNNT-based RMLSA algorithm can effectively improve the overall network performance.Compared with reference algorithms,the maximum optimization of WCES is 14%.4?The importance of the survivability of EONs is gradually increasing.How to provide a dynamic protection scheme that best matches the current network state under the premise of guaranteeing the network overall performance is one of the problems that need to be solved urgently to maintain the normal operation of the network.In this dissertation,a double-agent based RMLSA algorithm under the deep reinforcement learning framework(DA-DRL-RMLSA)is proposed.The corresponding RMLSA strategies are learned independently by the working agent and the protection agent respectively,and then strategies learned by these two agents are integrated in the controller to achieve a complete RMLSA policy supporting network survivability.The DA-DRL-RMLSA algorithm combines the previous network states,current network states and the possible future network states to select a working and protection RMLSA combination with the largest value of WCES.The simulation results show that the DA-DRL-RMLSA algorithm can improve the overall network performance by up to 23%and the network survivability by up to 16%,compared with the reference algorithms.