VAST Data, Dell, Versity and Spectra Logic are shining storage stars on TACC’s Horizon

TACC’s Horizon supercomputer is using VAST Data online storage, a Dell caching archive front end, Versity software, and Spectra Logic offline tape storage to hold its data.

TACC, the Texas Advanced Computing Center at the University of Texas in Austin has an upcoming Horizon AI and simulation supercomputer, which will be the largest academic supercomputer dedicated to open scientific research in the world. It will be the flagship of the U.S. National Science Foundation’s (NSF) forthcoming Leadership-Class Computing Facility (LCCF). Horizon is a 400 petaFLOP/s that will deliver a 10X simulation performance improvement and a 100X increase in AI capabilities compared to TACC’s existing Frontera system. Frontera, which was once #5 on the TOP500 list of the world’s fastest computers, uses 56 PB of DDN EXAScaler Lustre storage

VAST Data is supplying its AI OS as the data layer for Horizon. TACC says Horizon will have 400PB of local all-flash storage “delivering well more than 10TB/s of read/write bandwidth along with multi-tenancy and Quality-of-Service capabilities.”

Dan Stanzione, TACC’s Executive Director, said: “VAST has helped us streamline how we manage and access data across Stampede3 and Vista for more than two years, and we’re excited to extend that approach to Horizon as we unlock faster time-to-results for our researchers – less time moving data, more time running models, training AI, and publishing discoveries.”

VAST Data had previously won a 13 PB deal to supply storage to TACC’s Stampede3 supercomputer in July, 2023, and also supplied storage for TACC’s Vista system.

It says its AI OS “provides one consistent data experience across Stampede3, Vista and Horizon, so users and admins carry forward familiar workflows as capacity and performance step up – helping new communities onboard quickly.”

Read a VAST blog to find out more about VAST’s history at TACC.

Horizon will be supported by Ranch, an an exabyte-scale archive using Spectra Logic TFinity Plus tape libraries. Spectra is working with Dell and Versity to deliver the Ranch archive system, one of the highest-capacity archive systems Dell has ever delivered. There will be two 15-frame tape libraries with twenty LTO-9 drives that can support one exabyte of storage capacity. The system will be upgraded to add sixteen LTO-10 drives.

Versity’s ScoutAM software archive management system operates in a 2-tier, flash-to-tape scheme. It will run on 13 Dell PowerEdge R760 servers and automate data movement across five Dell ME5 storage arrays forming a 4PB flash tier, 16 PB of extended cache Dell ECS object storage, and the Spectra TFinity libraries. VAST Data’s all-flash platform provides the high-speed tier for active workloads for Horizon. Versity CEO Bruce Gilpin was asked about the VAST-to-Ranch data movement and said: “Right now this is all user initiated activity using GridFTP. Longer term we may add some tooling.”

As the Ranch archive ingests data from VAST, Versity’s ScoutAM then seamlessly handles the automated placement of that data into TACC’s long-term archive according to policy.

Kyle Lamb, Field CTO from VAST, said: “Both platforms were architected within the past decade and take advantage of state-of-the-art algorithms that unlock the full performance potential of modern hardware. By pairing VAST’s ultra-fast flash tier with ScoutAM’s exascale-ready archive management, TACC gets a clean, efficient architecture that delivers performance, scale, and simplicity without the baggage of legacy designs.”

Spectra CEO Nathan Thompson said: “With Horizon set to deliver a tenfold improvement in simulation performance and a 100x leap in AI capabilities, TACC needed a proven archival solution to match its vision and academic needs.”

He said the TFinity Plus system has “unmatched density, resilience, and track record in mission-critical environments.” Get a TACC Ranch archive system case study here. Get a Versity TACC case study here.

Bootnote

Globus says GridFTP is a highly performant protocol that moves the bits on the wire between source and destination storage systems. The GridFTP protocol, developed over 25 years ago now, was designed and built to manage ever increasing scientific data volumes across highly distributed, yet connected sets of systems. Read more here.