It’s a new record. Dutch researchers involved in high-performance computing have managed to achieve a speed of 6.55 million random read IOPS (Input/output operations per second) using a storage node that’s been designed by a company called Fungible. Fungible is a storage start-up based in California, and according to them, these test results “represent the highest recorded performance between a single server reading data and a single storage target”. This speed of 6.55 million IOPS is “almost double the previous best”.
“What we are achieving in the lab…can be deployed throughout the world…Ultimately, it is revolutionising the performance, economics, reliability and security of scale-out data centres,” said Pradeep Sindhu, CEO, and co-founder of Fungible.
According to the Block & Files report, this record was achieved by using Fungible’s FS1600 storage node. This storage node was powered by a data processing unit (DPU) which was also designed by Fungible.
More From This Section
The test was conducted by SURF and Nikhef jointly. SURF is an association of Dutch educational and research institutions, and Nikhed is a partnership between six universities and the Institutes Organisation of the Dutch Research Council. As per reports, Nikhef “is on the hunt for a fast and affordable data processing mechanism with a view to efficiently process the data flowing from experiments at CERN when the High-Luminosity LHC (HL-LHC) accelerator becomes operational in 2026”.
In the test conducted by these Dutch researchers, the FS1600 storage node, which was placed inside a 2U, 24-slot NVMe SSDbox with two F1 DPU controllers, was used together with a 64-core AMD-powered server over an NVMe-over Fabrics connection.
But the results they have achieved are not all. Fungible claims that this technology can be “scaled linearly” to deliver a whopping up to 300 million IOPS in a single 40 RU rack. The company said that the test results show that FS1600 can help decrease cost per IOPS, which helps improve the utilisation of storage media further as compared to the software-defined storage solutions that currently exist. This makes storage nodes like the FS1600 useful for all sorts of data-centric workloads.