This blog post is part 3 of a four-part series that explains how Nonvolatile Memory Express (NVMe), NVMe over Fabrics (NVMe-oF), and new storage-class memory (SCM) are changing the game for data centers. For a deeper dive, download the white paper “New Frontiers in Solid- State Storage”.
The pace of change in storage media technology continues to accelerate. Just a few years ago, 15k RPM spinning hard disk drives (HDDs) were the state of the art. Then came NAND flash and solid-state drives (SSDs), which replaced spinning drives for performance-sensitive workloads. Now we have new products such as Intel’s 3D XPoint and Samsung’s Z-NAND media, which are commonly classified as SCM, or persistent memory (PMEM). SCM can be connected either on the memory bus or on PCIe over NVMe.
These media technologies blur the line between memory and storage and can be used as either. Unlike NAND, some of the SCM technologies are byte addressable and can be used as memory. When deployed over NVMe (or NVMe-oF), these technologies can act as storage; on the memory bus, they can act as both memory and storage.
Because the expected media latencies (<2us) are lower than the network latencies (<5us), SCM will most likely be more prevalent on the host than on the network. However, SCM on a storage system could help accelerate metadata access and result in improvement of overall system performance. And with NVMe-oF providing low-latency access to networked storage, SCM can be potentially used to create a different tier of network storage.
SCM is approximately 10x faster than NAND flash, with better endurance, but it can be 5x to 8x more expensive than NAND-based SSDs. In some ways, this situation is similar to the time when NAND flash started getting mass produced about a decade ago. NAND flash started out at least 10x more expensive than the dominant 15K RPM HDD performance media of the time.
NetApp’s Plan for SCM
Most new low-latency media technologies require software architectures that can use them. Over the past 25 years, NetApp has learned how to deal with media changes and has adapted our products to different media as and when they become viable. We were an early adapter of consumer-grade SATA drives for use in enterprise storage. We were also an early adapter of flash with our NetApp® Flash Cache™ products. Later, when SSDs become prevalent, we adopted them to primary storage. Having the right software architecture makes all the difference.
History repeats itself. Like SSDs, most of the SCM technologies are very new and are going to be an order of magnitude more expensive than high-performance NAND. We expect to start using this low-latency SCM as a caching module in our All Flash FAS controllers. As the technology and media mature, we will move this SCM over to persistent storage. To get the maximum value of these SCM-based NVMe drives, the software has to be tuned. In some ways, this is a journey that every storage vendor has to go through, but we have been on this journey for at least the last five years as we helped develop NVMe standards and worked with leading partners to evaluate their SCM technology. As a result, we are confident our software architecture and designs are or will be ready for this kind of media as it matures and reaches enterprise readiness.
These new technologies are also creating a new ultrafast tier of host storage. This situation will require that many of the data management and resiliency capabilities that customers demand of their storage array will have to be offered in harmony on the host. It’s still early days, but keep watching: as the data management authority for the hybrid cloud, NetApp is committed to unlocking the value of your data with optimized performance and cost, whether it’s on the host, in networked storage, or anywhere else in your Data Fabric.
Visit our booth at Flash Memory Summit, August 4–10, 2017, and attend our keynote, “Creating the Fabric of a New Generation of Enterprise Apps,” on Thursday, August 10, 11:30 a.m. to noon.
Explore the implications of these new innovations in the other three blog posts in this series: