Researchers develop simulation technology for easy processing of big graph data

By Lim Chang-won Posted : April 23, 2021, 11:20 Updated : April 23, 2021, 11:20

[Gettyimages Bank]

SEOUL -- Researchers have developed graph processing simulation technology that can greatly reduce resources and time required to create and validate large-scale graph data processing algorithms. It enables the easy development of graph data-based services that have algorithms to calculate graph data.

The state-run Korea Advanced Institute of Science and Technology (KAIST) said its research team led by Kim Min-soo, a professor of computer science, has developed T-GPS (trillion-scale graph processing simulation) technology. Super-sized graphs can be processed with just one personal computer without the burden of storing data directly into a large cluster of servers

Researchers think the industrial ripple effect of T-GPS will be great. "With T-GPS technology, we can accurately calculate how the current algorithm will perform when graph data increases due to service operation by inputting small graph data secured in early service and algorithms in service to simulate," Kim said.

"It can also be applied to technologies that visualize only data parts accessed by participants without storing actual huge graph data in games and virtual reality software or calculate responses," the professor said.

Usually, the development of graph algorithms goes through a two-step process of generating and storing synthetic graphs and putting them back into memory in a graph processing engine to perform algorithm calculations. Large computational resources are required when graph data grows in size.

When performing graph data algorithm calculations, Kim's team said that T-GPS treats graph data that is not actually stored in the system as if it were stored, and the results of calculation can be obtained the same as when actual graph data is calculated. The calculation speed of algorithms was far faster than existing methods.

Kim's research was published at the online IEEE International Conference on Data Engineering (ICDE) on April 22, under the title "Trillion-scale Graph Processing Simulation based on Top-Down Graph Upscaling."
기사 이미지 확대 보기