Design And Performance Analysis Of High-Port-Count Large-Scaled Optical Switching System

With the continuous development of information services,the increasing data traffic has brought higher and higher requirements for networks' switching speed and capacity.In order to achieve sustainable development of future network infrastructures,the high-port-count high capacity fast optical switching system is a promising technology and gained significant attentions in the network research community.Meanwhile,as cloud computing and big data services become more and more prevalent,data centers(DCs)are playing crucial roles as the most fundamental computing platforms.The large scale optical switching system is also regarded as a promising candidate for implementing intra-DC networks.To this end,this thesis studies several key problems of the high-port-count large scale optical packet switching system,which includes analyzing its structure and scalability,evaluating its latency performance with different system configurations,and considering its application in the intra-DC scenario with a novel computing-task-oriented flow scheduling algorithm.Specifically,the contributions of this thesis can be summarized as following:Firstly,we analyze the structure and composition of the switching system,and discuss two scheduling mechanisms,namely the Round-robin and Long-queue-first scheduling.In order to quantitatively assess the system's performance under different scheduling approaches and with varying system parameters such as different port-count and speedup configurations,we built an OPNET-based simulation model and conducted extensive system evaluations.The results revealed the correlations among cost,complexity and delay performance of the system,which may provide some instructive clues for future practical application of the concerned optical switching system.Secondly,concerning the system's application for intra-DC scenario where many distributed-computing tasks may cause collections of correlated flows(Coflow)with particular deadline requirements,a more specific scheduling algorithm should be designed so that the network can take computing applications' particular flow transmission constraints into consideration and guarantee the tasks' performance.Hence we propose a Deadline-driven scheduling algorithm for the concerned optical packet switching system.By dynamically authorizing flows and precisely scheduling flow-classified optical packets,the Coflows' transmission time can be optimized to satisfy their deadline requirements as much as possible.Simulation results verify that our approach is effective and has obvious advantage over traditional long-queue-first scheduling method in terms of improved Coflows' completion ratio.