Mar 12 2019
The Open Compute Project (OCP) Global Summit is this Thursday and Friday (March 14-15) at the San Jose Convention Center, and NGD Systems will be there! You might wonder what a solid-state drive (SSD) company has to do with OCP, which was built around the concept of designing cheaper, more efficient vanity-free servers for massive data centers. The answer is simple: NGD Systems and its storage devices are more than just SSDs – they are computational storage devices, and computational storage devices are made to improve the performance of exactly the types of applications that run in OCP data centers – massively parallel, petabyte-scale workloads.
OCP was launched by Jonathan Heiliger at Facebook in April 2011 as a means to effect “open-source like” sharing of the designs for next-generation data center hardware. In a very real sense, OCP was a response by (what were at the time) second-tier cloud service providers who were looking for a way to compete with the purpose-built servers of Amazon AWS, Microsoft Azure, and Google, and the economies of scale and CapEx/OpEx reductions that they enabled. The concept was that by getting a variety of companies to collaborate around revolutionizing the datacenter in ways that “conventional” server vendors would not. This extended from new rack-level designs all the way down to I/O cards and storage tray designs, all with the concept of massively improving datacenter density, performance, and economy through the use of non-obvious ideas and approaches.
This concept of utilizing non-obvious ideas and approaches was exactly what drove the founders of NGD Systems to build our computational storage problems. We observed early on that data movement between storage devices and servers was increasingly becoming a significant performance issue for petabyte-scale applications. This position has been validated by studies showing that up to 50% of the energy in data centers is consumed moving data from storage to servers. To us, the obvious solution (which apparently was non-obvious!) is to NOT move the data from storage devices to servers, but to move (at least some of) the server processing capabilities into storage devices. This allows our devices to minimize data movement, improving application performance while reducing power consumption and costs. Facebook’s FAISS is but one of the applications that we have run on our computational storage devices, demonstrating significantly better performance than standard server architectures.
Last week we launched our newest product, the 16TB Newport U.2 NVMe® SSD. Newport is the world’s most advanced U.2 SSD, and is based on our world-class Newport computational storage platform. If you are going to be at the OCP Global Summit and would like to hear more about NGD Systems and what our computational storage devices can do for your petabyte-scale applications, contact me to set up a meeting. See you later this week!