CW Asia-Pacific

CW APAC: Buyer’s guide to NVMe storage

Konstantin Emelyanov - Fotolia

Tip

7 misconceptions about the NVMe protocol and NVMe technology

See what you need to know to avoid some of the common misperceptions about NVMe technology and making possibly expensive mistakes with your organization's NVMe architecture.

NVMe hasn't been around long, so it's not surprising there's confusion about it. At this point, even storage veterans have relatively little real-world experience with NVMe.

To set the record straight and possibly prevent you from making an embarrassing and costly mistake, here's a look at the real story behind seven widespread misconceptions about the NVMe protocol and NVMe technology.

1. NVMe is only useful with directly attached Peripheral Component Interconnect Express SSDs.

Actually, NVMe now goes far beyond functioning as a high-performance host controller interface for directly attached SSDs. It has evolved into a storage architecture and command set encompassing everything from small client SSDs to cloud-scale remote pools of SSDs.

"As deployment of disaggregated storage solutions using NVMe/TCP [NVMe over TCP] gain[s] adoption ... users will quickly realize that NVMe is much more than a direct-attached SSD standard," said Eran Kirzner, founder and CEO of Lightbits Labs, which offers a technology that separates compute and storage, enabling each to scale independently. "They will see they can have the performance of direct-attached from a networked NVMe/TCP storage solution over standard Ethernet."

2. NVMe-oF requires complex host bus adapters that support Fibre Channel, InfiniBand or remote direct memory access over converged Ethernet.

The NVMe standards committee recently standardized NVMe/TCP, which enables NVMe to be used over standard Ethernet networks with commodity Ethernet interfaces. "The amazing thing about NVMe/TCP is that it achieves performance -- including for latency and throughput -- close to that of other fabrics, but it doesn't require any special hardware or any modifications to the network infrastructure," Kirzner said.

3. We don't need the NVMe protocol; existing interface technologies are sufficient to get the job done.

That's just not the case in the era of big data analytics, AI and other cutting-edge technologies. Workloads are getting larger and more demanding, causing new problems for existing technologies.

"SATA and SAS interfaces were designed with spinning disks in mind, and they run into all sorts of bottlenecks when it comes to workloads which require fast access to storage," said Ian McClarty, president and CEO of data center services company PhoenixNAP.

4. There are few, if any, hardware NVMe RAID controllers.

NVMe has revolutionized storage and ushered in a new era of storage innovation.
Eran KirznerFounder and CEO, Lightbits Labs

One of the big misconceptions is that there are few hardware NVMe RAID controllers and that NVMe SSDs are mainly for caching tiers in software-defined storage or simple, nonredundant usage. "If configured properly, software RAID in Linux, using a large amount of NVMe SSDs, can outperform almost anything else on the market," said David Gustavsson, CTO and senior vice president of engineering at Axellio, a hyper-converged infrastructure and edge technology provider. "Software bottlenecks can quickly become a huge deterrent for adoption," Gustavsson added. "Understanding how these bottlenecks materialize and how to architect around them can pay off big time."

5. Using the NVMe protocol requires updating your entire infrastructure.

Not necessarily. The NVMe protocol can coexist with other transport protocols, and adopters may not need a new infrastructure to take advantage of NVMe on the front end for host connectivity, according to Greg Knieriemen, NetApp's director and marketing evangelist for storage systems and software. "For instance, NVMe and SCSI can concurrently run on the same SAN infrastructure, and NVMe protocol enablement is a nondisruptive software upgrade," he added.

6. NVMe is wildly expensive.

The prices of NVMe flash storage have dropped significantly lately and are pretty much comparable to SATA SSDs, PhoenixNAP's McClarty said. Capacity has also increased, making NVMe drives an option when building larger-capacity, all-flash storage systems. NVMe SSDs now make up more than half of all enterprise shipments, according to G2M Research. NVMe SSDs aren't as cheap as spinning disk SATA drives, he added, "but they are at a point where using SATA flash storage will no longer be considered."

NVMe SSD shipments

7. Most enterprise data centers don't need NVMe technology.

Completely false. "All data centers can benefit from the use of NVMe," according to Michelle Nowak, whose VPNMash blog covers privacy and other tech issues. "These systems enable organizations to reduce the number of physical servers and database application licenses, thus reducing the operating costs," she said.

Modern workloads -- such as AI, machine learning, real-time analytics and, generally, all workloads that must accommodate massive data sets -- require access to fast storage technology, McClarty said. Data center operators have two choices: They can tap into massive amounts of dynamic RAM (DRAM) or use NVMe drives emulating RAM pools through software. "Making use of such large DRAM capacity ... is often cost-prohibitive," he said. "That's where NVMe drives play a huge role in terms of usability and performance guarantees."

Don't get stuck

NVMe is a game-changing technology. "NVMe has revolutionized storage and ushered in a new era of storage innovation," Lightbits Labs' Kirzner said. Organizations that remain stuck on previous-generation storage will eventually pay the price in terms of both performance and enterprise competitiveness.

Article 2 of 3

Next Steps

Get to know NVMe protocol basics with these related terms

NVMe Technology in the Real World

Dig Deeper on Flash memory and storage

Get More CW Asia-Pacific

Access to all of our back issues View All
Disaster Recovery
Data Backup
Data Center
Sustainability and ESG
Close