Large-scale Graph Analysis Using the “Fugaku” Supercomputer

Summary

• 
  RIKEN, Institute of Science Tokyo, Fixstars Corporation, Nippon Telegraph and Telephone Corporation (NTT), and Fujitsu Limited have jointly conducted large-scale graph analysis using the supercomputer “Fugaku”.
• 
  “Fugaku” has been ranked first in the BFS (Breadth-First Search) category of the Graph500^* for 10 consecutive terms. Its score of 204.068 TeraTEPS is the first in the world to exceed 200 TeraTEPS.
• 
  Using 152,064 nodes (about 95.7% of the total) of “Fugaku”, they solved the breadth-first search problem for a super-large-scale graph consisting of about 8.8 trillion vertices and 140.7 trillion branches in an average of 0.69 seconds. In the latest measurements, succeeded in processing large-scale graphs that had never been seen before in the Graph500 ranking by using the memory-saving technology developed.

Background

Complex phenomena in the real world are often expressed as large-scale graphs (which show the relationships between data using vertices and branches), so high-speed graph analysis using computers is necessary.

By converting large amounts of data obtained through technology into graphs and processing them at high speed on computers, the development of new businesses that create new value is

High-speed graph analysis is expected to significantly contribute to promoting the Sustainable Development Goals (SDGs), particularly Goals 9 (Industry, Innovation and Infrastructure) and 11 (Sustainable Cities and Communities).

Solution

By achieving first place in the BFS category of the Graph500, ”Fugaku” has demonstrated that it is capable of performing regular calculations, which are often used in scientific and technological calculations but also highly capable of performing graph analysis, which is mostly irregular calculations.

This demonstrated the excellent versatility of ”Fugaku”, which can be used for various applications.

The joint research group is working to improve the scale of graph data that can be processed by “Fugaku”, and will accelerate performance improvements based on the data obtained in this measurement.

* Graph500: A project that ranks the TOP500 supercomputers based on their graph processing performance.