The National Computing Infrastructure’s 180,000 core, nine teraflop Gadi supercomputer is adding petabytes of new storage this year, under a deal signed with UK vendor SoftIron.
Image: NCI
The initial 12.5 PB of object storage will serve an Openstack cloud framework, and will act as active project storage for projects using the system.
The facility will be operating four distinct storage classes: the high-performance computing (HPC) storage, a facility-wide high-performance persistent storage platform, a cloud-focused object store, and a large-scale hierarchical storage management (HSM) based archival storage system.
SoftIron’s HyperDrive storage appliances will be backed by the Ceph open source storage framework.
Explaining what drove the need for new storage, an NCI spokesperson told iTnews: “Increasingly we are seeing the growth of edge computing with pre- and post-processing requirements being carried out as part of our cloud installation in support of the ‘big iron’.
“This equipment will complement our existing storage options and will sit between our fast parallel Lustre filesystems and our archival tier of storage.
"It will help round out the storage ecosystem of services for our researchers.”
The NCI spokesperson couldn’t provide an installation timeframe to iTnews, saying: “We currently only have the proof-of-concept equipment in place and are waiting for the delivery of the main equipment purchase.
"Production availability is subject to supply chain delays.”
The Gadi supercomputer came into service in November 2019, and in January 2020 replaced its predecessor, Raijin.
Gadi has already expanded from the 3200 nodes it had in January 2020 to more than 4000 nodes. At launch, it had 8 PB of operational disk storage.
NCI’s associate director of cloud services Andrew Howard said the NCI chose Ceph “for its maturity and robust capability as a unified, distributed storage system for emerging and non-traditional HPC capabilities."
Howard said Ceph offers "flexibility in its ability to offer object, block and file storage protocols in a single storage cluster, with the ability to scale virtually infinitely, as demands arise."
