A load balancing method and system for dynamic graph incremental computation

Abstract

The invention discloses a load balancing method and system for incremental computation of dynamic graphs. The method comprises the steps of utilizing degrees, initialization increment values and convergence times of vertices in the last iterative calculation process to generate a prediction histogram; after a graph data structure changes, initializing the state values and increment values of all the vertices in graph data; before new iteration calculation starts, utilizing the histogram and the degrees and initialization increment values of new vertices to predict convergence times corresponding to all the vertices; counting load conditions of all the vertices of each partition from slave nodes of the partition, sending the load conditions to a main node, dispatching and balancing the load of each partition by the main node, and returning decision results to each slave node; dispatching loads from migration partitions of the slave nodes to target partitions or receiving loads from overload partitions. Through the load balancing method and system for the incremental computation of the dynamic graphs, after the dynamic graphic data changes, the load condition of each partition in the new iteration calculation process can keep balanced, and therefore the effect of speeding up graph iterative calculation is achieved.

Description

technical field [0001] The invention belongs to the technical field of dynamic graph computing, and more specifically relates to a load balancing method and system for incremental computing of dynamic graphs. Background technique [0002] In the era of big data, more and more applications rely on the modeling and analysis of graph problems, and require real-time analysis and calculation on large-scale dynamic social network graphs. [0003] Incremental computing technology can effectively improve the processing performance of dynamic graphs. When the graph data is constantly changing, by storing the previous calculation results, when the input graph data changes, initialize the latest state value and incremental value of the vertices through the initialization method of incremental computing. Quantity value, the incremental value of vertex data is forwarded to the next iteration object as a message to support asynchronous incremental calculation, which can effectively reduce...

AI Technical Summary

Problems solved by technology

[0005] In view of the above defects or improvement needs of the prior art, the purpose of the present invention is to provide a load balancing method and system oriented to incremental calculation of dynamic graphs, thereby solving the inability to solve the problem of load inclination during calculation running in current load balancing technology problems, as well as the technical problems of additional communication overhead and synchronization overhead caused by load tilt detection during optimized runtime

Images