TORONTO – Solid state drives (SSDs) that use the Non-Volatile Memory Express (NVMe) interface haven't been adopted as quickly as expected, but one storage startup sees the advent of dual port functionality as an opportunity to make inroads the enterprise storage market.
Zivan Ori, CEO and co-founder of E8 Storage, told EE Times in a telephone interview that he sees the industry shifting toward NVMe on both the client and enterprise side. "There's no longer a premium associated with NVMe," Ori said.
What's changed, according to Ori, is the ability to share SSDs in the same enclosure by leveraging the interface. "That's the market we are addressing," he said.
E8 is an Israeli company with research and development in Tel Aviv and sales and marketing in Santa Clara, Calif. Founded in late 2014, E8 began engaging customers in April 2016 and its first product was on the market in December. E8 recently opted to put HGST's Ultrastar SN200 PCIe SSDs into its NVMe enclosures.
Generally speaking, said Ori, SSDs are becoming faster, but customers are grappling with the choice of using SSDs as local storage in their servers to get high performance or put those SSDs in all flash arrays in the SAN to gain manageability and provisioning. He said inevitably, they go with the first option. However, it means each SSD only belongs to that server, their content can't be shared between servers, and they must overprovision, which leads to inefficient use of capacity.
In the second scenario, the dual controller becomes the bottleneck in the all-flash array. E8 is solving that problem, said Ori, with its scale-out architecture using NVMe SSDs and 40GE to 100GE networks. Centralized NVMe makes sense for those using local SSDs, he said, and applications such as SQL and NoSQL databases, real-time analytics, and high performance computing, while also enabling new applications, such as machine learning, Internet of Things databases and analytics, and real-time application performance monitoring. “That's why we capitalized on NVMe,” Ori said.
E8 has chosen to select off-the-shelf components, said Ori, and while it's selected Western Digtal-owned HGST most recently, it's happy to work with any SSD vendors depending the drive's features. Although he couldn't provide many details, E8 is currently collaborating with Intel on 3D Xpoint.
Walter Hinton, senior director of enterprise devices marketing at Western Digital, said the adoption of NVMe has taken longer than expected from a vendor perspective. “We had some growing pains within a couple of operating systems environments,” he said, but Western Digital now has NVMe across its product lines.
E8 is addressing the challenge of dual controllers becoming the bottleneck in an all-flash array with its scale-out architecture using dual port HGST NVMe SSDs and 40GE to 100GE networks.
E8 has selected HGST's second generation NVMe SSD, Hinton said, which really represents a combination of development from HGST, Western Digital and SanDisk. Hinton himself comes from Virident, which was acquired by HGST. He said Western Digital has been able to quickly merge the SanDisk roadmap for SSDs into the broader organization.
Earlier generations of HGST's NVMe SSD were single port devices that lent themselves to scale-out types of workloads such as MySQL, said Hinton. “With the introduction of dual port NVMe, which is what E8 is using from HGST, you know see the opportunity to bring NVme into storage arrays," he said. "It creates a different set of latency and IOPs characteristics.”
Hinton anticipates there will be a lot of activity around dual port NVMe SSDs. It's been a year since front-loading NVMe servers first appeared, and now they are available from all of the major server vendors. He said storage arrays using NVMe SSDs can tackle big data analytics workloads that traditional storage couldn't deal with. “We will see more arrays that have NVMe on the inside.”
But while NVMe adoption is picking up, it's by no means going to supplant other interfaces completely, said Hinton, and Western Digital, like other vendors, is segmenting its SSD product lines by interface. “SATA has a long, long life and has the performance to meet our needs," he said. "The SAS market continues to grow.” He said NVMe at the high end is growing rapidly right now. “We think it's on the ramp that it should be at this point.”
Hinton said the NVMe market will segment just as SSDs have in general for the enterprise. There will be a “Ferrari of NVMes” that will have the highest performance and lower latency, but customers will also have the option for an NVMe SSD that is more comparable to SATA but a little more expensive. “Customers will have more choice.”
Jeff Janukowicz, research VP at IDC for SSDs and enabling technologies is seeing an uptake in NVMe adoption as well, noting that while NVMe SSDs only accounted for 15 percent of the enterprise SSDs shipped in 2016, the forecast is for them to snag 60 percent by 2020. He said last year marked an inflection point for NVNme where it is moving beyond niche applications.
Janukowicz said when SSDs first came out, it made sense to look at them like a hard drive and still see a bump in performance, even in the early days of PCIe, but to really take advantage of the inherent benefits of the technology, such as high performance and low latency, much of the ecosystem needed to change, he said, and the standardization of the NVMe protocol has taken time.
PCIe offers a lot of flexibility, such as the number of available and different power envelopes, said Janukowicz, while dual port NVMe SSDs will contribute to their adoption in storage infrastructure as it has typically leveraged SAS dual controllers. “It's a good example of the evolution of the ecosystem."
—Gary Hilson is a general contributing editor with a focus on memory and flash technologies for EE Times.