Add to My Interests
Remove from My Interests
Building a persistent high-performance-compute-ready data storage platform with a capacity of over 8 petabytes and throughput of over 140 gigabytes per second presents many challenges, both from a design and operational perspective.
The National Computational Infrastructure (NCI) is Australia’s national high-performance research computing facility, managing the Southern Hemisphere’s most integrated supercomputer and filesystems, delivering high-quality computational and data services to researchers in three national science agencies and nearly thirty of Australia’s universities.
NCI is home to Raijin, a 57,000-core Sandy Bridge Intel® Xeon® system, ranked at #24 in the HPC Top 500 list on debut, with 30 petabytes of high-performance storage hosting one of Australia’s largest data catalogues of nationally and internationally significant research data.
To meet the growing demand for additional capacity for multi-petabyte data collections, NCI commissioned an additional 8PB+ large-scale global persistent file storage platform known as gdata3. The gdata3 storage system is based on Intel's Enterprise Edition for Lustre®, with scalable storage modules provided by NetApp® E-Series arrays (E5600) and EF-Series (EF550) all-flash technologies.
This presentation will discuss the design solution used to deliver the gdata3 storage platform that enables researchers using NCI's facilities to access and collaborate on collections of 1-3PB in size from both the high-performance 1.2 petaflop supercomputer and OpenStack cloud systems.
Topics discussed will include hardware selection and logical design, component performance validation and benchmarking, data migration plus system resiliency through backup and recovery.
Customer Presented Sessions
Do Not Sell My Personal Information