Research On Elastic Compute Method Oriented To Trajectory Stream

At the present stage,the 14-day trajectory checking has become a common means of normalized epidemic prevention and control,and the real-time processing based on trajectory big data is effectively used in the scenarios of spatio-temporal concomitant monitoring and close contact identification of suspected cases in epidemic prevention and control.At the same time,as the construction of smart cities enters a new phase,the mode and form of integration of trajectory big data computing and urban governance is undergoing significant changes,and the trajectory analysis performed in offline scenarios is no longer satisfied with the status quo.The ever-changing applications require the"fresh" value of trajectory data to be fully explored,and there is an urgent need to break through the real-time management and accurate query means of large-scale trajectory data,which is of great significance to enhance the capability and digital intelligence of urban governance.Trajectory streams have natural characteristics such as real-time,dynamic and spatiotemporal skew,so traditional distributed computing means such as solidified resource allocation and spatial partitioning cannot achieve high throughput and low latency trajectory stream computation.Existing research shows that in the task scenario of real-time processing of traj ectory stream data,resilience issues such as load balancing and cluster level scaling are the crux of the problem.With the aim of enforcing the fundamental principles,this thesis delves into the problem of resilient computing for trajectory streams and proposes:(1)Random distribution-based elastic compute method for trajectory stream:Random distribution does not perceive the spatio-temporal characteristics of the trajectory flow and is therefore naturally resistant to skewing.This method achieves dynamic,accurate and efficient trajectory stream join operations by distributing the task load evenly among the job nodes through a random distribution scheme generated by a matrix model.(2)Hash distribution-based elastic compute method for trajectory stream:Hash distribution is fully aware of the spatio-temporal characteristics of the trajectory streams and is therefore able to mine the distributional properties of the trajectory streams.This method combines the advantages of hexagonal grid layout addressing and learned indexing in terms of time and space complexity,and effectively works for low overhead and low latency trajectory stream range queries.(3)Dynamic index-based elastic compute method for trajectory stream:Indexing is a key technology for offline big data processing,but it is difficult to be applied to realtime scenarios due to frequent updates and other problems.This method aims to adapt the traditional index structure to be compatible with the streaming computing model,thus enabling stable and high-performance continuous Top-k trajectory similarity queries.In summary,this thesis presents research work on resilient computing and query processing for trajectory streams.With the objective of investigating the integration of load balancing and resource self-adaptation mechanisms in real-time trajectory stream processing systems,the thesis explores theories and methods to support the complete life-cycle data management of trajectories and builds a prototype system.