The Shared Hierarchical Academic Research Computing Network, SHARCNET, is a consortium of Canadian academic institutions providing Advanced Research Computing (ARC) resources, including compute and storage, to academic researchers across Canada. As one of the country's national high-performance computing (HPC) sites, SHARCNET supports the diverse long tail of the ARC community, from bioinformatics and astrophysics to humanities and social sciences.
Unlocking Innovation for Canadian Research
Building a stable, scalable, and AI-ready future for Canadian researchers.
Industry
Education
Use Case
Artificial Intelligence
Video
Overview
Background
For 25 years, SHARCNET had relied on a Lustre-based file system, a solution John Morton, Director of Technology, SHARCNET describes as “the race car that is very fast but requires constant attention to keep running.” While performing well at its peak, the system’s lack of stability and reliability created persistent issues that could take down the entire compute cluster, frustrating both the administration team and the thousands of researchers who depend on the resource.
Outcome
Since deployment, VAST has delivered on its promise of stability, allowing the SHARCNET team to focus on innovation rather than fire-fighting. The shift to VAST has been transformational for the technical team. The solution is so reliable it's become nearly invisible to the administrative team. Morton relayed a telling anecdote from a recent system administrator meeting: "Has anyone looked at VAST because I haven't. Everyone in the meeting was just shaking their head going ‘nope, haven't looked at it, haven't seen any tickets.’ There's no reason to even log into the system because it's just taking care of things."
VAST directly benefits AI researchers ensuring stability with high-IOPS and high-file-count workloads. While the performance of VAST meets their needs, the key benefit for these users—who generate large numbers of small files and high IOPS—is the platform's resilience. VAST is designed to keep running reliably without being destabilized by the intensive I/O patterns typical of large AI research.
Has anyone looked at VAST because I haven't. Everyone in the meeting was just shaking their head going ‘nope, haven't looked at it, haven't seen any tickets.’ There's no reason to even log into the system because it's just taking care of things.
John Morton
Director of Technology, SHARCNET