Flash storage in networking infrastructure needs to focus on reliability, quality and data retentions

5 mins read

The infrastructure to support and grow connectivity is constantly evolving, and encompasses telecommunications, data communications and data centres. The processing and storage applications in that infrastructure stretches from base stations to subscriber lines, through a hierarchy of routers and switches. As the amount of digital traffic continues to expand, the need for fast and reliable storage only increases.

Comms infrastructure requirements differ significantly from those found in the enterprise and consumer sectors. In enterprise the focus is on speed and low latency, typified by redundant arrays of independent disks (RAIDs) standardised on PCI Express and SATA interfaces. This is a large market and has many “me too” product offerings, which is also true for the consumer sector, where the overriding requirements are cost and capacity, less performance and longevity.

With comms infrastructure’s the emphasis is on reliability, quality and data retention. Of course, cost is a factor, but so too are continuity of supply and lifecycle support. Failing devices add significantly to the cost as is evidenced by the need to monitor the health of storage devices and, if necessary, perform failure analysis.

The ability to operate across an extended temperature range and support for diversity in the interfaces used are also requirements. The later includes also a need to support legacy interfaces with state-of-the-art Flashes that are available in the market. Most manufacturers focused on the enterprise or consumer sectors aren’t able to meet these demands.

Legacy support and TCO

The specific challenges found in the comms infrastructure stem from the viewpoint of the “meta-system”, in which every sub-system is part of a larger overall system. Storage is an important element of this, but represents a small percentage of the total system. However, despite being small, storage devices form part of a critical path, and any failure will have significant consequences.

Storage solutions for comms applications need to be designed to minimise the total cost of ownership (TCO)

over the system’s entire lifetime, which is consequently significantly longer than comparable solutions in the enterprise or consumer sectors.

Elements of the system may need to be replaced or upgraded over that lifetime, which from a storage point of view can involve migration to the latest interface standard. Suppliers in this sector need to offer support for new and legacy interfaces, as well as the long-term availability of storage solutions.

Axel Mehnert is responsible for marketing, product strategy, and business development at Hyperstone

The usage may be more diverse compared to other sectors. Random accesses and read performance together with read-disturb management is important for code and operating system storage. Random write performance and, coinciding small write amplification are vital to ensure endurance for logging of small amounts of incremental data or configuration updates. Rapid burst performance is required when writing DRAM-crash dumps to Flash in the event of a power outage. A sophisticated wear-levelling and garbage collection is needed when these different types of use share the same physical NAND Flash components.

In the enterprise domain, NVMe is emerging as a high-capacity, high-performance storage solution, which uses the PCI Express interface to connect non-volatile storage media. However, as a result of this format it also comes with high power requirements and elevated system costs. Coupled with high power and system costs, NVMe can be subject to comparatively lower power fail robustness due to increased user and management data-caching, all of which makes it unsuitable for Communications applications.

The embedded or eUSB module format is based on, and compatible with, the USB interface standard and compatible to USB 2.0 and 3.1. It integrates a controller to provide a managed NAND solution and can be a suitable replacement other form factors such as SSD or eMMC. Using NAND memory, the controller can handle wear levelling and provides a flexible storage alternative. eUSB modules are available focusing on industrial and embedded applications using different Flash technologies including MLC, pSLC and SLC. Capacities usually range from 2 to 128Gbyte with performances reaching 160Mbyte/s. The flexibility to properly dimension storage system according to storage requirements helps to keep power and system cost low while offering endurances and reliability higher than other formats, including SSD and NVMe.

As an alternative storage solution, eUSB modules are a compelling proposition supporting full-, high- and super-speed transmissions as defined by the USB 3.1 standard, making it up to 30 times faster than USB 2.0 offerings, but with backward compatibility. Self-Monitoring, Analysis and Reporting Technology (SMART), allows the module’s health to be closely monitored. At the heart of the modules is the U9 USB 3.1 Flash memory controller and management technology developed by Hyperstone. Using hyMap Flash translation layer (FTL) and hyReliability firmware it maximises endurance and data retention, and provides robust power and fail-safe functionality. Data integrity is ensured under all circumstances for the whole lifetime.

Featuring an embedded 32-bit processing core with an instruction set optimised for Flash memory management, along with an AES 128 and 256 encryption engine, the U9 family offers 16 general-purpose I/O and an API with SDK to support the development of customer-specific firmware extensions (CFEs). Figure 1 shows a block diagram of the U9 Flash controller.


Health monitoring has become integral to Flash memory management in the Communications infrastructure. Working in conjunction with a controller like the U9, it enables parameters such as total spare blocks, block erase counts and ECC/CRC errors to be captured at any time, as well as checking the status of global wear levelling and bad block management.

All of this data is accessible through the hySMART utility, using a GUI to access and decode the ATA/vendor-specific data. The availability of C++ source code for the main functions, allows customers to incorporate SMART features at the system level. In addition, the data can be correlated and used for lifetime estimation helping customers evaluate their TCO.

An important part of calculating the TCO involves understanding the storage’s endurance. With Flash-based non-volatile memory this is related to how the physical medium maps to the host’s logical storage patterns. This mapping is referred to as the Flash translation layer, or FTL; all Flash memory media uses the FTL to map where data are stored.

The physical storage locations in the NVM are arranged in blocks, pages and in some cases sub-pages. Mapping with finer granularity (sub-pages) enables the NVM to be used optimally; however, this puts a greater burden on the Flash controller algorithm, so it becomes a trade-off for some controller manufacturers. Hyperstone’s hyMap technology uses sub-pages by default, which helps with another feature known as the write amplification factor (WAF). This is a figure that indicates how much additional information is written to the Flash for every byte of data; a lower WAF is therefore better.

Achieving a low WAF depends on many things but starts with the granularity, the page and the block size, and the underlying Flash technology: does it use single-level or multi-level cells, for example (SLC or MLC)? The WAF can also be impacted by where the mapping information is stored: in internal or external DRAM. As a rule, a lower WAF and finer mapping can boost endurance by a factor of as much as 100 especially for use-cases with frequent small random accesses such as boot drives for example. Determining the WAF for a given write budget can help in selecting the most suitable Flash technology; using hyMap, it may now be possible to use MLC or pSLC instead of SLC.

Above: the figure shows a block diagram of the U9 Flash controller


Storage is a small but integral part of the larger systems that comprise comms infrastructure; a fault in any sub-system could result in service failures. NVM has become the default media for code and data storage across the entire infrastructure, meaning service providers rely on suppliers being able to provide long-term support for new and legacy interfaces.

The way NVM is used in communications demands an approach that isn’t just tailored to the needs of the enterprise or consumer sectors, where lifecycles are short it’s about endurance, reliability and TCO.