New Requirements, New Architectures: The Evolution of HPC Storage

Over the last 10 years, the expectations for storage solutions have changed at HPC sites, and the trends that brought those changes will continue. In the old days, uptime was measured in weeks, because peak performance was the only metric of merit. Today, we have AI/ML workloads taking significant compute time, edge computing and distributed datacenter strategies bringing WANs into our workflows, and cloud-based compute to somehow factor into our plans. As our workflows evolve, we are recognizing the metrics of uptime and overall utilization of computer clusters as more important than peak performance.