chris - Fotolia


NVMe Zoned Namespace lowers costs and improves performance

The new zoned storage interface capitalizes on NAND flash capabilities to let host software communicate with NAND storage to better align workloads with SSD storage.

The NVM Express consortium recently ratified the NVMe Zoned Namespace Command Set specification. It defines an interface for enabling host software to communicate with non-volatile NAND flash storage.

Zoned Namespace (ZNS) divides the NVMe namespace into independent zones that support specific data or workload types. It moves data management to the host software to improve control of I/O and data placement. With NVMe Zoned Namespace, applications can exploit NAND's inherent architecture, leading to lower storage costs and better performance, particularly for hyperscale workloads.

The challenges of traditional flash storage

An SSD comes with several limitations. Data is written sequentially and must be erased before being rewritten. Data is written at the page level but erased at the block level. An SSD supports only a limited number of these program/erase (P/E) cycles before memory cells start to fail. If not properly managed, these issues can negatively affect SSD endurance and shorten an SSD's lifespan. For this reason, most SSDs include a flash translation layer (FTL) that handles life-extending tasks such as garbage collection and wear leveling, helping to address the SSD's shortcomings.

Unfortunately, the FTL management layer can negatively affect throughput and increase latency and costs, particularly with zettabyte-scale workloads. For example, the FTL is constantly moving data around as part of the ongoing P/E cycles, resulting in write amplification and possible performance degradation. An SSD must also be overprovisioned to accommodate these management processes, and it requires adequate dynamic RAM (DRAM) to support them.

Zone technologies may eventually trickle down to data centers of all sizes, making flash storage a viable option for more organizations.

Because of the FTL's impact, especially at scale, companies such as Alibaba, Microsoft, NetApp and Western Digital have been working on a zoned-based approach to SSD management that eliminates the FTL overhead and increases SSD efficiency. As part of this effort, the NVM Express Technical Working Group developed the ZNS interface to provide an open standard that aligns SSD internals with the support host.

The ZNS interface extends the NVMe protocol to provide a logical address space that enables host software to direct I/O traffic to specific zones, similarly to how shingled magnetic recording (SMR) zones work for HDDs. With the NVMe Zoned Namespace interface, workloads are better aligned with their SSD storage, improving performance and lowering costs.

Moving toward zone-based storage

The ZNS interface incorporates many of the principles of Open-Channel architectures used to control SSD devices. An Open-Channel SSD exposes its logical block address (LBA) space directly to the host so it can manage data placement, I/O operations and other processes. The LBA is divided into chunks made up of logical blocks that the SSD's sector size defines. The chunks must be written sequentially, but they can be read randomly. The host directs each workload to specific chunks, resulting in better resource use and performance.

Storage vendors and hyperscale organizations have been adopting Open-Channel architectures to address FTL limitations. However, this has led to a variety of specialized implementations, without a common standard. NVM Express hopes to change this situation with the ZNS interface.

The interface divides the LBA into zones, similar to the Open-Channel chunks. Each zone is treated as an isolated namespace. Data is written sequentially within the zone and must be erased before it can be rewritten. This approach is achieved without the management overhead required of the FTL. In addition, different data and workload types can target specific zones to provide more predictable use patterns and better use storage resources.

How ZNS works
ZNS decreases write amplification and overprovisioning, reducing costs and increasing throughput and latency predictability.

The ZNS interface aligns with the Zoned Block Command (ZBC) and Zoned-device ATA Command (ZAC) interfaces, which support SMR zones on SAS and SATA HDDs, respectively. As with ZNS, each SMR zone is an independent unit. Data must be written sequentially and erased before the zone can be rewritten.

Aligning the ZNS interface with ZBC and ZAC makes it easier to incorporate ZNS in existing infrastructures that support SMR. For example, SMR-enabled device mappers and file systems should be able to work with ZNS with few, if any, changes. This alignment also makes it possible to implement single zoned-based storage that incorporates both HDDs and SSDs.

ZNS and hyperscale workloads

The NVMe Zoned Namespace interface promises to offer several benefits. For example, write amplification is reduced because zones better align with the SSD's physical geometry, eliminating the need to constantly move, erase and rewrite data with every operation. At the same time, ZNS reduces storage overprovisioning, in part because there's less write amplification, but also because data is stored more efficiently through zones.

Another benefit of ZNS is that it moves data management to the host, reducing the need for DRAM, while lowering administrative overhead and storage costs. Lower overhead leads to better throughput and latency rates. More efficient data placement and I/O controls further enhance performance. These benefits translate to greater drive endurance, helping to reduce storage costs even more.

The NVMe Zoned Namespace effort targets hyperscale organizations, such as cloud providers and global data centers. Hyperscalers must manage their storage as efficiently and cost-effectively as possible, making ZNS an ideal fit. Several storage vendors are developing ZNS-enabled SSDs to meet these needs.

Western Digital has been at the forefront of this effort, participating in the development of the ZNS interface and through its own Zoned Storage Initiative, which takes a zone-based approach to delivering storage that incorporates both HDDs and SSDs. Western Digital has already released SMR-enabled HDDs and offers a ZNS-enabled development SSD to select customers, with more SSDs planned.

SK Hynix's has demoed a 2 TB NVMe ZNS SSD.

Radian Memory Systems, Samsung and SK Hynix are actively working on ZNS-enabled SSDs. Radian and SK Hynix have demoed products that meet ZNS specifications, and Samsung offers a drive that can support ZNS implementations. And it's not just storage vendors embracing ZNS. SANBlaze Technology, a storage protocol testing specialist, recently announced it's adding ZNS support to its validation and test platform.

The emerging world of ZNS

Now that the NVMe Zoned Namespace specification has been ratified, other vendors are expected to move into the market. Nevertheless, the interface has a long way to go before there's widespread adoption, so its impact and success are yet to be seen. Open-Channel provides an indication of its potential, but it won't be until hyperscale organizations implement ZNS that we'll understand its full impact and how SSD storage might evolve as a result.

Even so, ZNS represents an important step in addressing the challenges that come with FTL-based SSDs. Zone technologies may eventually trickle down to data centers of all sizes, making flash storage a viable option for more organizations.

Next Steps

Storage as the Driver of Change: Rethinking Data Infrastructure

Making NVMe Drives Handle Everything from Archiving to QoS

How Zoned Namespaces Improve SSD Lifetime, Throughput, and Latency

Dig Deeper on Flash memory and storage

Disaster Recovery
Data Backup
Data Center
and ESG