The world’s first 1.5 ExaFLOPS supercomputer is about to be paired with an equally trailblazing storage subsystem that offers about 700 Petabytes (PB) of storage, 75 TB/s of throughput, and 15 billion input/output operations per second of performance.
The Frontier exascale supercomputer is being assembled by the Oak Ridge Leadership Computing Facility (OLCF) and is set to go online in late 2021.
To keep pace with Frontier’s exascale computing prowess, OLCF has announced the Orion storage subsystem, which it claims is the world’s fastest.
“To the best of our knowledge, Orion will be the largest and fastest single file POSIX namespace file system in the world,” said Frontier’s Input/Output Working Group lead at OLCF, Sarp Oral.
Pushing the envelope
Orion is made up of three tiers, a NVMe storage tier, a hard disk storage tier, and a flash-based metadata tier.
First up is the NVMe storage tier that uses 5400 SSDs for a total of 11.5PB, and offers a peak read-write speed of 10TB/s.
Then there’s the hard disk tier with 47700 magnetic recording disks with a peak read speed of 5.5TB/s, a peak write speed of 4.6TB/s that add another 679PB of storage.
Finally, there’s the metadata tier, which is made up of 480 NVMe SSDs with a combined capacity of 10PB.
“Orion is pushing the envelope of what is possible technically due to its extreme scale and hard disk/NVMe hybrid nature,” noted OLCF’s Group Leader of High-Performance Computing Storage and Archive Group, Dustin Leverman.
As for the software, Orion will use the open source Lustre parallel file system, which also powers several other HPCs, along with the ZFS logical volume manager.
Via Tom’s Hardware