Dutch High Performance Computing (HPC) scholars have successfully achieved a random read speed of 1600 million IOPS using a storage node developed by Fungible. According to Fungible, a California-based storage company, the test results represent the highest performance ever recorded between a single data reader server and a single storage target, nearly double the best ever. "What we do in the lab...can be incorporated globally...Ultimately revolutionizes the performance, economics, reliability and security of scalable data centers," said Pradeep Sindhu, director and co-founder of Fungible . Reporting on the development, Blocks & Archives claims that the record was set using the company's FSXNUMX storage node powered by a Data Processing Unit (DPU) developed by Fungible itself.
Download storage logic
The test was carried out jointly by SURF, an association of Dutch research and educational institutions, and Nikhef, a partnership between the Organization of Institutes of the Netherlands Research Council and 6 universities. According to reports, Nikhef is seeking a fast and affordable data processing mechanism to efficiently process test data at CERN when the High Illumination LHC Accelerator (HL-LHC) becomes operational in 1600. In the Dutch test, the FS2 storage node, housed in a 2-slot 1U NVMe SSD enclosure with 1600 FXNUMX DPU controllers, was used with an AMD XNUMX-core server via an NVMe-over Fabrics connection. Fungible claims that the technology can be scaled linearly and deliver up to XNUMX million IOPS in a XNUMX RU rack. The company says the results show that the FSXNUMX helps reduce cost per IOPS, further improving storage media utilization, compared to existing software-defined storage solutions, making them useful for all kinds of applications. workloads.