As high-performance computing consumes more and more data and the demand for shorter turnaround times increases, the speed of data transfer becomes an increasingly important bottleneck. Now in an article published in IEEE Transactions on Network and Service Management, researchers at the University of Samsara and the University of Missouri announced the development of an algorithm that increases data transmission speeds up to one and a half times in powerful data centers.
The algorithm is based on a special routing method. First, the user enters four basic parameters: bandwidth, data transfer speed in Kbps, cloud storage and price. The algorithm then uses a shortest path search algorithm to provide data transmission at the necessary quality and speed. This âneighborhood method,â as they call it, is an approach that scientists have already developed and successfully published in the context of network virtualization.
Scientists claim that the data transfer method is particularly relevant for high-precision calculations in basic science and applied research. One of the scientists behind the research, Andrey Sukhov, explains how the algorithm could add value to major projects like the International Thermonuclear Experimental Reactor in France – or even CERN’s Large Hadron Collider.
“They calculate the tasks in laboratories scattered around the world, inquire at computer centers,” Sukhov said. âThey also need to exchange both text information and high-resolution streaming video online. The technology we are proposing will help them do that.
The researchers hope to use this algorithm in the near future to create an experimental application to study combustion reactions. If successful, the implementation would allow scientists around the world to access the project remotely. More broadly, it would also allow remote access to the powerful Sergey Korolev supercomputer.
About the paper
The document referred to in this article is titled “A constrained path scheme for managing virtual network services. “It was published in August 2018 by IEEE Transactions on Network and Service Management and written by Dmitrii Chemodanov, Flavio Esposito, Prasad Calyam and Andrei Sukhov.