Mar 06 2019
NGD Systems has announced the general availability of their Newport Platform for computational storage. The Newport Platform uses In-Situ processing where the compute is inside the host to remove the need to move data to memory for processing. That removes a processing bottleneck, and greatly facilitates Intelligent Edge deployments, Those deployments are also facilitated by these NVMe devices’ small form factor, large capacity and low power consumption. Those advancements are significant enough that partners are able to sell the devices just on those TCO benefits.
The Newport Platform, which updates and replaces NGD Systems’ previous platform, uses the world’s first 14nm SSD Controller, which enables the In-Situ processing and thus eliminates the moving datasets. The In Situ processing brings a Hadoop workload into the drive itself and lets it run locally. It gives a 35 per cent plus performance improvement, while using half of the CPU cores.
“This is the next generation of storage,” said Nader Salessi, NGD Systems’ CEO and co-founder. “The uniqueness of it is its combination of large capacity and low power consumption. Any time you have data sitting in the storage, it’s always more efficient to do it inside the host. If the storage device is very small, it’s more efficient to move it to the host. But as soon as you have more than a few TB of data, moving it to the host takes too much time, and uses a lot of power. So by moving the compute inside the host instead, it provides distributed processing in parallel.”
“This responds to the massive amounts of data being generated, and the need to assess it, by bringing compute resources into the storage device itself, using its processor, without using a host resource,” said Scott Shadley, Principal Technologist and VP of Marketing at NGD Systems.
NGD Systems began its operations and computational storage development in 2013, and brought their first units to market – FPGA prototype solutions – in 2017. Last year, they updated them with new FPGA, and enhancements to features, form factor and capacity. Along the way they acquired 26 patents relating to computational storage, with 10 more pending. Their own initiatives coincided with broader developments in the industry. SNIA formed a new TWG around computational storage standard in 2018, in response to market demand, to drive further customer adoption. Shadley is the chair, and 27 companies have now joined the TWG. The last Flash Memory Summit also held its first full session on computational storage.
“We have talked with end customers about what they wanted from storage,” Shadley said. “They have all iterated on existing products in the SSD space. So we looked at a ground-up approach at what customers want.”
With the Newport platform, NGD Systems is now delivering the world’s first ASIC-base design, and is stressing that the Newport platform’s computational storage capabilities addresses several key firsts.
“We built an ASIC-based controller that provides cost, performance and scale, and is the first with 14nm ASIC-based NVMe SSDs,” Shadley said. “We crammed 16 TB into that 2 and a half inch footprint, as the controller capability allows us to get that dense. That means that this is the first NVMe SSD solution to provide native support up to 32TB of Storage in a 2U form factor, which solves capacity problems.”
He also stressed that it does this while reducing power consumption.
“We need to be power-efficient while doing this, and this is the first NVMe solution to require under 10 watts of active power with over 16TB of storage per drive.”
Shadley also said that this is the first NVMe SSD to offer five nines of QoS in a high capacity SSD.
“The industry talked for years about having a million IOPs of throughput, which no one can use, but latency is very important,” he stated.
“We will also be coming out with additional form factors in EDSFF, replacing the M.2 form factor, as those platforms become available,” Shadley added. The likely timeframe for this is Q3.
NGD Systems sold direct from their inception in 2013, and since then, they have added a small number of channel partners.
“For the channel, the trick is that it is a very flexible platform, and we have been selling it in what we call a ‘walk-run’ strategy,” Salessi said. “We have shipped the SSDs as a high performance NVMe storage device to channel customers who will resell it to any customer who can use it. They want the density it offers, while they investigate the potential for turning on computational power later on. The big deal now is the high capacity and low power.”
At this stage, the channel is a small number of larger system integrators, but they will eventually expand that.
“There’s nothing that prevents the platform from being a global-scale product,” Salessi stated.
A full channel program is likely in the Q2 or Q3 time frame.