The U.S. Department of Energy’s (DOE) Argonne National Laboratory (ANL) has released a request for a new storage cluster for its supercomputer fleet. According to the Datacenter Dynamics resource, the project could cost $15–$20 million.
We are talking about creating a storage system that will provide the capacity and performance necessary to support the operation of existing NPC complexes, as well as future supercomputers. It is noted that the Argonne Leadership Computing Facility (ALCF) has deployed several high-performance parallel file systems to process data generated by researchers and engineers. These are, in particular, two Luster systems with a capacity of 100 PB with a throughput of 650 GB/s. They both use Infiniband HDR interconnect.
Image source: DOE
The new storage system will have a capacity of 400 PB. Requirements include IOPS performance of up to 240 Mbps, peak throughput of 6 TB/s, POSIX compatibility and the ability to simultaneously mount up to 30 thousand nodes. The vendor must provide support for five years.
It is expected that the platform will be used by the Aurora supercomputer, which ranks second in the current TOP500 ranking with a speed of 1,012 Eflops. In addition, the Polaris HPC complex will have access to the storage system: its peak performance is about 44 Pflops.
The designed storage system must provide “the reliability and scalability required for the next generation of HPC and AI.” The platform must be delivered to the contractor by the second or fourth quarter of 2025, if an additional six months allows the introduction of new technologies.