Communication method and apparatus applied to computer cluster

Patent metadata

Field	Value
Publication number	US-12348328-B2
Application number	US-202218572373-A
Country	US
Kind code	B2
Filing date	Oct 10, 2022
Priority date	Oct 11, 2021
Publication date	Jul 1, 2025
Grant date	Jul 1, 2025

Abstract

A computer cluster is configured to perform distributed computing on graph data, the computer cluster includes a first computer and a second computer, the first computer is configured to process a vertex in a first vertex set in the graph data, the second computer is configured to process a vertex in a second vertex set in the graph data, the first vertex set includes a first source vertex, the second vertex set includes a first destination vertex corresponding to the first source vertex. A communication method includes: generating information of the first source vertex; determining the second computer based on pre-recorded mapping relationship information, where the second computer is configured to process information of the first destination vertex; and sending a first communication message to the second computer, where the first communication message includes the information of the first source vertex.

First claim

What is claimed is:

1. A communication method applied to a computer cluster, wherein the computer cluster is configured to perform distributed computing on graph data, the computer cluster comprises a first computer and a second computer, the first computer is configured to process a vertex in a first vertex set in the graph data, the second computer is configured to process a vertex in a second vertex set in the graph data, the first vertex set comprises a first source vertex, the second vertex set comprises a first destination vertex corresponding to the first source vertex, and the method is applied to the first computer,
the method comprises:
generating information of the first source vertex;
determining the second computer based on pre-recorded mapping relationship information, wherein the second computer is configured to process information of the first destination vertex; and
sending a first communication message to the second computer, wherein the first communication message comprises the information of the first source vertex, so that the second computer updates the information of the first destination vertex based on the information of the first source vertex.

2. The method according to claim 1 , wherein the generating information of the first source vertex comprises:
generating a graph embedding vector corresponding to the first source vertex; and
deleting some or all of zero elements in the graph embedding vector corresponding to the first source vertex, to obtain the information of the first source vertex.

3. The method according to claim 1 , wherein the first computer exchanges a communication message with another computer in the computer cluster in a round-robin manner.

4. The method according to claim 1 , wherein the first vertex set further comprises a plurality of destination vertices, and the plurality of destination vertices respectively correspond to a plurality of source vertices;
the method further comprises:
receiving a second communication message, wherein the second communication message comprises information of the plurality of source vertices; and
processing the information of the plurality of source vertices in parallel by using a plurality of threads.

5. The method according to claim 1 , wherein the computer cluster performs a graph learning task based on the graph data.

6. A communication apparatus, wherein the communication apparatus is a first computer in a computer cluster, the computer cluster is configured to perform distributed computing on graph data, the computer cluster further comprises a second computer, the first computer is configured to process a vertex in a first vertex set in the graph data, the second computer is configured to process a vertex in a second vertex set in the graph data, the first vertex set comprises a first source vertex, the second vertex set comprises a first destination vertex corresponding to the first source vertex, and the communication apparatus comprises a processor executing instructions stored in memory to implement:
generating module, configured to generate information of the first source vertex;
determining module, configured to determine the second computer based on pre-recorded mapping relationship information, wherein the second computer is configured to process information of the first destination vertex; and
sending module, configured to send a first communication message to the second computer, wherein the first communication message comprises the information of the first source vertex, so that the second computer updates the information of the first destination vertex based on the information of the first source vertex.

Assignees

• Alipay Hangzhou Inf Tech Co Ltd

Inventors

• Qin Wei
• Yu Jiping
• Zhu Xiaowei
• Chen Wenguang

Classifications

• H04L67/10 Primary

  in which an application is distributed across nodes in the network (software deployment G06F8/60; multiprogramming arrangements G06F9/46) · CPC title

• H04L12/185 Primary

  with management of multicast group membership · CPC title