Even with massive computing nodes and the high speed of networked devices, the spindle drive limitations of conventional RAID storage systems render their performance inadequate to handle the parallel IO workloads of HPC applications. Distributed file systems such as Lustre, which is commonly used for HPC storage, often become a bottleneck and diminish overall performance. Replacing RAID storage with a top-tier all-flash array system can easily make the difference in reaching maximum performance. An all-flash array that can handle high IO loading matched with a Metadata Server (MDS) is perfect for improving performance by 2-3 times as compared to a disk array.
AccelStor NeoSapphire™ all-flash arrays use the HPC industrial standard InfiniBand FDR 56G/s interconnection and adopt the Remote Direct Memory Access (RDMA) protocol to eliminate network bottlenecks. AccelStor’s proprietary flash management software, FlexiRemap, effectively handles parallel heavy write workloads. What’s more, enterprise businesses can reap significant saving in terms of electricity usage and rack cabinet space by using AccelStor all-flash arrays.
High Performance Computing (HPC) is now widely used for tasks ranging from numerical analysis, weather prediction and artificial intelligence to machine learning and computational chemistry, among others. Leveraging advancements in parallel multi-core CPU/GPU computing farms and high-speed interconnected networks, HPC processes data and delivers it to distributed storage systems extremely quickly.