On November 18, the 18th Green Graph 500 and the 23rd Graph 500 rankings were published at the Supercomputing Conference (SC) held in St. Louis. The graph processing accelerator DepGraph, developed by professional team from National & Local Joint Engineering Laboratory for Big Data Technology and System at the School of Computer Science and Technology of HUST, ranks first in the Green Graph 500 ranking. With a performance-power consumption ratio of 6234.32MTEPS/w, it also ranks first in the Graph 500 in terms of stand-alone performance - reaching 884.361 GTEPS and 997.491 GTEPS for SSSP and BFS applications, respectively. (All the official names from http://grid.hust.edu.cn/zlxz/sysLOGO.htm)

It is reported that Graph 500 and Green Graph 500, as the most authoritative lists in evaluating the graph computing performance and performance-power consumption ratio of supercomputers worldwide, were first released at the SC in 2010 and 2012 respectively. They mainly evaluate current popular graph computing, which can fully reflect the computing efficiency, memory access performance and communication performance of supercomputers when supporting graph computing, directly reflecting the ability of supercomputers to process graph data.

The graph computing team is guided by Associate Professor Zhang Yu. The team members include HUST PhD candidates Zhao Jin (class of 2017) and Qi Hao (class of 2021).Also contributing are Jiang Xinyu and Yang Yun, two master students from the class of 2020. After years of in-depth research, breakthroughs have been made in several key technologies of graph processing accelerator and graph computing system software. DepGraph, the graph processing accelerator used by the team for ranking, can effectively regularize and speed up the transfer of graph vertex states, decouple data dependencies, ensure efficient joint data access and load balancing, so that the upper graph algorithm can make full use of the lower parallel computing resources, making the performance of two typical graph algorithms SSSP and BFS reach 884.361 GTEPS and 997.491 GTEPS respectively and the performance-power consumption ratio reach 6234.32 MTEPS/W, both ranking first in the world.

According to Zhang Yu, graphs can effectively express the relationship between things and are the basis of data analysis and application. Important applications such as artificial intelligence can process more complex and large-scale data by using the processing method of graph data, increasing the efficiency and accuracy. Therefore, graph computing, as the core technology of the next generation of AI, has been widely used in many fields such as medicine, education, military affairs, finance and others. For example, financial anti-fraud analysis has become a familiar concept to us. However, the characteristics of irregular data access, complex dependency and low computational access memory ratio of graph computing cause traditional computing’s architecture great challenges when supporting graph algorithms.

The School of Computer Science and Technology has previously won many awards in the field of graph computing. In August, it won the 2021 Graph Challenge Championships. This was the first win for a Chinese team at this event. In October this year, Yuntu, the world's first high-performance graph computing system for concurrent graph analysis tasks, won the gold medal in the Industry Track of the 7th China International “Internet+” College Students Innovation and Entrepreneurship Competition.

Source: School of Computer Science and Technology, Huazhong University of Science and Technology

Written by: Huang Bochuan

Edited by: Li Ruiyao, Scott, Peng Yumeng