- NVMe Storage
- Enterprise NVMe
For many years, HPC was characterized by large single purpose clusters with expensive, proprietary hardware tuned to specific applications. Over the last decade a good deal of work has been done to create more general purpose multitenant clusters that can support multiple concurrent users and different applications. The ADS1000 was designed to support this movement. It works hand in hand with parallel file systems like Lustre in order to quickly warm-up client datasets like those used by the world’s most powerful supercomputers as ranked by the TOP500 list (www.top500.org).
Apeiron supports high-speed server to server messaging using the Apeiron storage fabric so that no external switching, interface cards, or alternative protocols are required. The messaging uses direct memory access using the same queues and buffers designed for NVMe over Ethernet storage in order to get single digit microsecond transfers using a standard Open Fabric Alliance (OFA) software stack (See www.openfabrics.org). This makes the ADS1000 particularly useful in FinTech applications or for scientific analysis as in genomic sequencing.
NVMe over Ethernet is ideally suited to Big Data financial applications such as pre- and post-trade analytics since its I/O performance eliminates large numbers of unneeded servers and provides for dramatic query performance improvements.
By providing different types of SSD media all in one storage environment, high performance media like Intel’s Optane™ SSDs can be used for block cache or high-performance storage pools of hot data while standard NAND SSDs can be used to store warm data all managed by resource managers like YARN or MESOS. This allows applications to tailor their use of storage media to the data needs. For more information, please read the Hadoop Reference Architecture document.