IMAGE: Trillion-scale Graph Processing Simulation
(T-GPS) Technology
view more 

Credit: KAIST

A KAIST analysis crew has developed a new expertise that permits to course of a large-scale graph algorithm with out storing the graph in the primary reminiscence or on disks. Named as T-GPS (Trillion-scale Graph Processing Simulation) by the developer Professor Min-Soo Kim from the School of Computing at KAIST, it may possibly course of a graph with one trillion edges utilizing a single pc.

Graphs are broadly used to signify and analyze real-world objects in lots of domains corresponding to social networks, enterprise intelligence, biology, and neuroscience. As the variety of graph functions will increase quickly, creating and testing new graph algorithms is changing into extra vital than ever earlier than. Nowadays, many industrial functions require a graph algorithm to course of a large-scale graph (e.g., one trillion edges). So, when creating and testing graph algorithms such for a large-scale graph, a artificial graph is often used as an alternative of a actual graph. This is as a result of sharing and using large-scale actual graphs may be very restricted because of their being proprietary or being virtually not possible to gather.

Conventionally, creating and testing graph algorithms is finished by way of the next two-step method: producing and storing a graph and executing an algorithm on the graph utilizing a graph processing engine.

The first step generates a artificial graph and shops it on disks. The artificial graph is often generated by both parameter-based era strategies or graph upscaling strategies. The former extracts a small variety of parameters that may seize some properties of a given actual graph and generates the artificial graph with the parameters. The latter upscales a given actual graph to a bigger one in order to protect the properties of the unique actual graph as a lot as potential.

The second step masses the saved graph into the primary reminiscence of the graph processing engine corresponding to Apache GraphX and executes a given graph algorithm on the engine. Since the dimensions of the graph is just too giant to slot in the primary reminiscence of a single pc, the graph engine usually runs on a cluster of a number of tens or tons of of computer systems. Therefore, the price of the traditional two-step method may be very excessive.

The analysis crew solved the issue of the traditional two-step method. It doesn’t generate and retailer a large-scale artificial graph. Instead, it simply masses the preliminary small actual graph into major reminiscence. Then, T-GPS processes a graph algorithm on the small actual graph as if the large-scale artificial graph that needs to be generated from the actual graph exists in major reminiscence. After the algorithm is finished, T-GPS returns the precisely similar end result as the traditional two-step method.

The key thought of T-GPS is producing solely the a part of the artificial graph that the algorithm must entry on the fly and modifying the graph processing engine to acknowledge the half generated on the fly because the a part of the artificial graph truly generated.

The analysis crew confirmed that T-GPS can course of a graph of 1 trillion edges utilizing a single pc, whereas the traditional two-step method can solely strategy of a graph of 1 billion edges utilizing a cluster of 11 computer systems of the identical specification. Thus, T-GPS outperforms the traditional method by 10,000 instances when it comes to computing assets. The crew additionally confirmed that the pace of processing an algorithm in T-GPS is as much as 43 instances quicker than the traditional method. This is as a result of T-GPS has no community communication overhead, whereas the traditional method has a lot of communication overhead amongst computer systems.

Prof. Kim believes that this work may have a giant affect on the IT trade the place nearly each space makes use of graph information, including, “T-GPS can significantly increase both the scale and efficiency of developing a new graph algorithm.”


This work was supported by the National Research Foundation (NRF) of Korea and Institute of Information & communications Technology Planning & Evaluation (IITP).

-About KAIST

KAIST is the primary and high science and expertise college in Korea. KAIST was established in 1971 by the Korean authorities to teach scientists and engineers dedicated to the industrialization and financial development of Korea.

Since then, KAIST and its 64,739 graduates have been the gateway to superior science and expertise, innovation, and entrepreneurship. KAIST has emerged as one of the vital modern universities with greater than 10,000 college students enrolled in 5 faculties and 7 colleges together with 1,039 worldwide college students from 90 international locations.

On the precipice of its semi-centennial anniversary in 2021, KAIST continues to try to make the world higher by means of the pursuit in schooling, analysis, entrepreneurship, and globalization.

Disclaimer: AAAS and EurekAlert! are usually not liable for the accuracy of stories releases posted to EurekAlert! by contributing establishments or for using any data by means of the EurekAlert system.