NVMe: The storage underneath 5G and edge computing

Sold-state drives using the Nonvolatile Memory Express (NVMe) architecture support computers that include wireless routers, IoT and sensor network gateways, video cameras, and industrial PCs. Solid-state drives (SSDs) connect to hosts over PCIe, which can even run over network fabrics such as Ethernet. SSDs connected using NVMe and PCIe can attain higher data rates than other buses such as SATA and USB.

Where does NVMe and the computing equipment it supports fit into the overall 5G network architecture? To find out, 5G Technology World asked Hiep Pham, business development director at Virtium.

5GTW: With the transition from 4G to 5G communications underway, which industries are building out the infrastructure to take advantage of 5G’s wide coverage, huge user capacity, high throughput, and low latency over private networks?
Pham: As the 5G transition takes hold, it will involve connections that are not just people-to-people or people-to-machine, but machine-to-machine. We’ll see an infrastructure built out for industrial automation, which will leverage artificial intelligence (AI), and machine learning (ML) to improve their processes, efficiency, and production throughput. Autonomous vehicles will take advantage of AI and machine learning to enhance customer experience and safety. Defense and military applications where fast, precise and deterministic results are required will also use AI and ML. Additionally, Smart Cities will rely on 5G for public needs such as transportation, traffic management, parking, and public health and safety. On the consumer front, 5G speeds will enhance the mobile and virtualized gaming experience.

5GTW: How are edge-computing products being designed to take advantage of 5G, and where in the network will those computing products reside?
Pham: Edge computing is a high-level term that defines moving computing out of data centers, closer end points — that is, the devices positioned throughout the network edge. It demands fast responses at the end points and has to process and transfer huge amounts of data through the network, so 5G provides the necessary bandwidth, performance, and efficiency to meet those demands. There are, however, limitations on how fast data are stored and processed. The current trend is to distribute the storage and data processing based on importance and real-time requirements.

Edge computers such as wireless routers, gateways for IoT and sensor networks, video AI cameras for security and safety applications, and industrial PCs for manufacturing robotics and automation reside very close to network endpoints. Additionally, they’re at the edge of the network to handle data from multiple sources and interfaces. To handle different interfaces and data packets, the applications can be downloaded and configured, like with Network Function Virtualization (NFV), products. To handle many possible applications, edge-computing-based products need to have the flexibility of memory and storage to handle the application’s requirements of speed, low latency, and capacity. Let’s not overlook that. Due to what’s demanded of them, many of these applications require industrial-grade storage and memory.

5GTW: What are the advantages of different locations, such as at cell towers, base stations, 5G cores, and data centers?
Pham: The key advantage of different computing locations is the fast response versus cost. Cell towers are the first and closest point from the end devices, so latency is lowest there. But, placing computers at the cell towers increases cost, occupies the limited space, and demands that the equipment be designed to operate in a harsh, outdoor environment. Most base stations are situated very close to the end points so the latency increases only minimally. Some real-time applications can reside at the base stations although the cost, space, and, in some cases, the harsh environment could present a deterrent. 5G cores and data centers are farther away from the end devices and therefore experience the highest latency. The total equipment costs will, however, be lower. 5G cores can handle some partial real-time applications but most likely the data centers will handle all background computation and long-term storage.

5GTW: We hear a lot these days about NVMe storage in both the enterprise and industrial space. What does NVMe bring to the 5G party relative to more-established standards like SATA and eUSB in terms of reducing latency?
Pham: NVMe’s advantages over SATA and eUSB include lower latency, higher speeds, and greater flexibility. Reduced latency will enable many new applications, particularly for those that require significant I/O throughput out to the point of storage. NVMe also offers more software-enabled configurations, which will help to reduce latency of the responses and end results. Additionally, NVMe offers better security and a variety of network interfaces (through NVMe over Fabrics) to handle a huge variety of applications. Of course, it does require more power to operate and, therefore, generates more heat, which needs to be mitigated through careful system design.

5GTW: What challenges and opportunities do designers focused on 5G face regarding storing and protecting the mass of data networks collect? What should they consider essential in addressing those demands?
Pham: 5G, AI, ML, and data center technologies are evolving toward distributed communication and computing architectures. They require higher speed to transfer data from devices to data centers and more processing power at the edge. Partial AI processing can take place at or near the end devices. The key challenge for designers is to ensure that the deployed products are flexible and can handle the future applications. System designers will ask “What future applications and what capacity of storage do I need?”

As CPUs from Intel, NXP, and many others get more powerful, there’s an opportunity to build computer boards that can be downloaded with specific software and configured to operate in a particular application. We have seen some of these products — virtualized Open Radio Access Networks, White box, Network Function Virtualization, and Software Defined Wide Area Networking. These are the 5G-related virtualized network products based on a computer platform whose functions are downloaded and updated through software. In industrial automation and robotics, we’re seeing many custom industrial PCs using this model. To handle many possible applications, they need the flexibility of memory and storage performances to handle the applications’ requirements of speed, low latency, and capacity. Many of these PCs will need industrial-grade components.

5GTW: What would surprise most people about the relationship of 5G communications and industrial-grade storage?
Pham: Cellular communications can be broken into two connected paths: the fronthaul and backhaul. Open RAN network architectures open include a mid-haul transport as well. As the cellular system transitions to 5G, the front haul part will be changed significantly. Many components of the front haul like the radio, distributed and central units will be deployed in a constrained environment or outdoor environment, which includes the impact of very limited space, severe heat, variation of temperature, and humidity. This is where industrial-grade memory and storage products can deliver and perform consistently. Additionally, many industrial-grade storage modules have extended lifetime, due to how their NAND Flash memory is selected and built.

5GTW: Can engineers take advantage of NVMe in applications beyond network computing? For example, in IoT devices? How?
Pham: The NVMe architecture and interface present many advantages in both hardware and software. For hardware, it offers more capacity, lower latency, and faster speed. It will be the design choice for many applications that place a premium on performance. For Industrial automation and robotics, the faster speed and lower latency is critical. For video surveillance, the camera with AI capability can take advantage of the capacity, speed and low latency. For software, NVMe offers many advanced features and options including namespaces, which are used when a storage virtual machine is configured with the NVMe protocol. One or more namespaces are provisioned and connected to an NVMe host, and each can support various block sizes. In an environment with many IoT devices, for example, a gateway can take advantage of namespaces to organize and store data for each individual IoT device.

5GTW: What do system designers need to consider when implementing NVMe storage in terms of power, heat, or signal integrity? Does 5G make designing for these issues easier or more difficult? How?
Pham: NVMe requires more power and generates more heat that other technologies. The much-faster speeds can also cause signal integrity issues. For power, a designer can select the NVMe module with fewer PCI Express, commonly called PCIe, lanes — full capacity would be four lanes per standard module. Reducing PCIe lanes will reduce speed and at the same time reduce power consumption. Regarding heat, there are NVMe modules built with heat sinks that can help to reduce heat. As with heat, reducing the number of PCIe lanes also lowers the power consumption. The circuit design and component layout of the board interface to the NVMe SSD must be considered thoroughly and designed to carefully handle both signal level and timing signals. This is part of the electrical timing and impedance requirements for a higher-speed circuit design anyway, but it bears repeating when dealing with such powerful technologies as NVMe and 5G.

Related articles: