Asynchronous message sending method and system based on graph computing engine

Abstract

The invention relates to the field of graph data and specifically relates to an asynchronous message sending method and system based on a graph computation engine. The asynchronous message sending method based on the graph computation engine comprises the steps of S1, in each partition, sending messages received by a current vertex to a next vertex along the edge of the current vertex according to a predetermined rule until a preset condition is satisfied; S2, synchronizing the messages received by the next vertex in each partition; and S3, returning to carry out the step S1 if a termination condition is not satisfied. According to the asynchronous message sending method and system based on the graph computation engine, through adoption of an asynchronous message sending mode, information switching times among the partitions is reduced, network communication resources are saved, and the processing efficiency is improved.

Description

technical field [0001] The invention relates to the technical field of graph data, in particular to an asynchronous message sending method and system based on a graph computing engine. Background technique [0002] Large graphs are well suited for a variety of graph analysis and mining applications, such as social network analysis, pattern recognition, website classification, and social media marketing, among others. However, as the volume of graphics that needs to be processed becomes larger and larger, a single server is already overwhelmed. Therefore, a series of large-scale distributed parallel framework diagrams came into being. The distributed parallel frame graph is to divide a large graph into multiple sub-graphs, and then store these separately, so multiple sub-graphs can be run in parallel during calculation, greatly improving computing efficiency, and can be applied to large-scale real-world graphics problems. [0003] Graph cutting methods are divided into edge ...

AI Technical Summary

Problems solved by technology

Based on the existing graph data storage structure, the information of vertices and edges is stored separately, which increases the communication overhead; and the lack of index information in the storage structure greatly reduces the calculation rate

[0006] In addition, the existing synchronous message sending mode based on synchronous parallelism (BSP) requires frequent information exchange between each subgraph, which consumes a large amount of network communication resources, seriously affects the performance of the system, and cannot adapt to some large Real-World Graphics Computing

Images