"For SciNet, NVMesh is an extremely cost-effective method of achieving unheard-of burst buffer bandwidth", stated Dr. Daniel Gruner, chief technical officer, SciNet High Performance Computing Consortium. "By adding commodity flash drives and NVMesh software to compute nodes, and to a low-latency network fabric that was already provided for the supercomputer itself, NVMesh provides redundancy without impacting target CPUs. This enables standard servers to go beyond their usual role in acting as block targets - the servers now can also act as file servers."
Based in Toronto, SciNet, Canada's largest supercomputer centre, serves thousands of researchers in biomedical, aerospace, climate sciences, and more. Their large-scale modeling, simulation, analysis and visualization applications sometimes run for weeks, and interruptions can sometimes destroy the result of an entire job. To avoid interruption SciNet implemented a burst buffer - a fast intermediate layer between the non-persistent memory of the compute nodes and the storage - to enable fast checkpointing, so that computing jobs can be easily restarted. SciNet had deployed the Spectrum Scale (GPFS) shared parallel file system on their spinning disk system, but at scale, as individual jobs become larger, checkpointing may take too long to complete, making the calculation difficult, or even impossible to carry out.
Using Excelero's NVMesh in a burst buffer implementation, SciNet created a peta-scale storage system that leverages the full performance of NVMe SSDs at scale, over the network - easily meeting SLA requirements for completing checkpoints in 15 minutes, without needing costly proprietary arrays. With NVMesh, SciNet created a unified, distributed pool of NVMe flash storage comprised of 80 NVMe devices in just 10 NSD protocol-supporting servers. This provided approximately 148 GB/s of write burst (device limited) and 230GB /s of read throughput (network limited) - in addition to well over 20M random 4K iOPS.
Emulating the "shared nothing" architectures of the Tech Giants, SciNet's NVMesh deployment allows them to use hardware from any storage, server and networking vendor, eliminating vendor lock-in. Integration with SciNet's parallel file system is straightforward, and the system enables SciNet to scale both capacity and performance linearly as its research load grows.
"Mellanox interconnect solutions include smart and scalable NVMe accelerations that enable users to maximize their storage performance and efficiency", stated Gilad Shainer, vice president of marketing at Mellanox Technologies. "Leveraging the advantages of InfiniBand, Excelero delivers world leading NVMe platforms, accelerating the next generations of supercomputers."
"In supercomputing any unavailability wastes time, reduces the availability score of the system and impedes the progress of scientific exploration. We're delighted to provide SciNet and its researchers with important storage functionality that achieves the highest performance available in the industry at a significantly reduced price - while assuring vital scientific research can progress swiftly", stated Lior Gal, CEO and co-founder at Excelero.