Building End-to-End NVMe over Fabric Infrastructure for
HPC
SessionStorage
Presenter
Event Type
Exhibitor Forum
TimeWednesday, November 15th4:30pm -
5pm
Location503-504
DescriptionThe goals of non-volatile memory (NVM) storage for high
performance computing (HPC) generally involve minimizing
storage latency, especially when the NVM storage use
case is local and dedicated SSDs for burst buffers or
parallel file systems, or pooled and shared SSDs for
parallel file system or object storage acceleration.
Common high-performance SSD interconnection technologies
include Serial Attached SCSI (SAS) and PCI Express®
(PCIe®), yet even PCIe alone is no longer sufficient to
meet the performance requirements of present and future
HPC architectures or Storage Class Memory (SCM)
technologies.
During June 2016, The NVM Express® (NVMe®) organization published the NVMe over Fabrics 1.0 (NVMeoF) specification defining a common architecture supporting the NVMe block storage protocol over a range of networking fabrics. Commercial off the shelf (COTS) products are becoming available enabling end-to-end NVMeoF infrastructures and one reference architecture will be presented demonstrating next-generation latency, throughput, and bandwidth for high-performance storage.
During June 2016, The NVM Express® (NVMe®) organization published the NVMe over Fabrics 1.0 (NVMeoF) specification defining a common architecture supporting the NVMe block storage protocol over a range of networking fabrics. Commercial off the shelf (COTS) products are becoming available enabling end-to-end NVMeoF infrastructures and one reference architecture will be presented demonstrating next-generation latency, throughput, and bandwidth for high-performance storage.
Presenter
