Design And Implementation Of Graph Computing System Based On Multi-GPU

With the rapid development of computer-related industries,especially the emergence of mobile Internet,the data to be processed by computers is not only getting larger in scale,but also more and more complicated in structure.Graph is a common data structure,by abstracting individuals as vertices and relationships between individuals as edges in the graph,a lot of data is stored in the form of graphs.Graph computing is to discover the relationship of these large-scale and complex-structured graph data,and analyze,calculate it.With the help of graph computing,people can solve many difficult problems in daily life.In response to this problem,this paper proposes and implements a single-machine graph computing system TGraph based on a multi-GPU environment.By constructing virtual graphs in a multi-GPU environment and choosing appropriate design patterns for each iteration,TGraph can make full use of the parallel capabilities of the GPU.The main tasks included in this article are as follows:1.TGraph proposes a virtual graph calculation method in a multi-GPU environment.By abstracting the memory in the system,the virtual graph is shared as read-only for each GPU.2.TGraph proposed a dynamic selection strategy of pull-push calculation method.This dynamic strategy can dynamically select an appropriate calculation mode according to the load of the calculation during the calculation process,so that the TGraph can make full use of the parallelism of the GPU.Finally,by comparing with CPU-based graph computing system and GPU-based graph computing system,the performance of TGraph graph system was evaluated in detail.Experiments have also verified the performance scalability of TGraph on multiple GPUs.