Research On Complex Optical Networks In Support Of Diverse Computing Paradigms

Driven by cloud computing and edge computing,the number of computing applications has increased dramatically in the past 10 years.It can be predicted that with the development of communication technology and the further enhancement of the information width and depth,more new applications relying on computing will emerge and update quickly,which will generate more computing scenarios and will constantly challenge the existing networks.The number of computing applications running on the whole network is increasing,the amount of computation is increasing,the computing applications raise higher and higher requirements on collaboration and completion time,which can be concluded as the trend of"Diverse Computing Paradigms".In order to deal with this trend,the network structure should be studied based on the characteristics of complex traffic patterns of computing tasks.Meanwhile,the high density and dispersion of computing tasks make the network show the characteristics of complex network.To apply traditional routing algorithm directly in complex network can easily cause network congestion.In order to ensure the efficient transmission of computing tasks,it is necessary to study the routing algorithm.In addition,facing the diverse needs of massive computing tasks,a reasonable task scheduling and coordination mechanism need to investigated.Specifically,the main work of this paper can be summarized as follows:Firstly,in order to evaluate how well can the network support computing tasks more scientifically,this paper defines two new network evaluation metrics:mode completion time and mode communication length.Furthermore,based on the optimization objective of minimizing the average mode completion time,we propose a new network construction method called CoPa by uitilizing genetic algorithm in this paper.The simulation experiments are carried out under different communication patterns and their combinations.The network performance of CoPa can greatly exceed the most advanced network construction strategy,which shows that CoPa can deal with the challenges of the network structure caused by the superposition and mixing of massive computing tasks.Secondly,the network constructed by CoPa presents the complex network characteristics,so we focus on designing efficient routing strategy in complex network scenarios in this paper.We propose GH routing algorithm which combines the structure characteristics of scale-free network and the dynamic information of nodes.It selects the optimal path with the minimum weighted evaluation value.By building a large-scale network simulation platform,we verify that GH routing can not only achieve large traffic capacity but also small average end-to-end transmission time of packets.In addition,we deploy the GH routing algorithm in the data center network,which verifies its efficiency and practicability.Thirdly,because of the unbalanced spatial-temporal distribution of massive concurrent computing tasks and the need for extensive cooperation in multiple regions,we design the computing job scheduling mechanism in multi-Datacenter networks which is called MuReCo.Specifically,MuReCo integrates the scheduling and routing with both the computing resources and network resources considered,and it makes good coordination between multiple datacenters.Through the simulations in the real Internet service provider network,we can find out that MuReCo reduce the job completion time of computing tasks greatly under both uniform and non-uniform load distributions.Finally,we build an experimental platform to demonstrate the setting and adjustment mechanism of diverse computing patterns on the optical inter-edge/cloud datacenter network.It shows the feasibility and necessity of the collaboration between edge datacenters and cloud datacenters.