What is an all-flash array (AFA)?

An all-flash array (AFA) is a type of external data storage system that exclusively uses flash memory hardware media (that is, solid-state drives (SSDs)) for persistent data storage. The AFA is also called a solid-state storage disk system or a solid-state array (SSA).

Unlike spinning-disk hard-disk drives (HDDs), SSDs contain no moving parts and offer improved speed and durability. Compared to HDDs alone, all-flash storage systems—composed of SSDs or hybrid systems composed of both SSDs and HDDs—enable faster data access while improving processing speeds and reducing CPU strain. 

Flash storage ranges in both storage capacity and form factor. Some SSDs, like the kind found in USB flash drives, are smaller than dimes. And while early SSDs could hold only a small amount of data, as the technology progressed in accordance with Moore’s Law, that storage capacity multiplied.

Today, because of SSD technology, you can balance a terabyte of data (or more) on the tip of your finger. Due to their compact size and robust, solid-state construction, SSDs are ubiquitous in electronics like laptops, tablets, mobile phones and gaming consoles. 

Beyond consumer-grade applications, SSDs and AFAs have also had a revolutionary impact on enterprise storage. Traditionally, organizational data centers have relied on HDDs configured as network-attached storage (NAS) or storage area networks (SANs), or both, for data storage and retrieval. That is until advancements in solid-state storage made all-flash arrays a cost-effective alternative and ushered in several performance improvements suitable for high-performance applications.

Solid-state drives are typically preferable over hard disk drives. However, hybrid arrays that configure SSDs and HDDs within the same chassis enable vendors to retrofit existing systems by replacing only a portion of the fixed media with flash. These days, AFAs or hybrid arrays, are the leading choice for both NAS and SAN solutions at scale.

Flash memory is a type of nonvolatile memory storage that uses semiconductor transistors to store and retrieve data. Data stored in flash memory is persistent and nonvolatile, meaning that once data is written to a flash memory unit, the unit can and will retain that data even if powered off.

Flash memory technology comes in two varieties differentiated by the type of logic architecture used to manage the memory chip’s operations:

• NOR flash: This type of flash memory uses “NOT OR” Boolean logic gates to read and write data. NOR flash is fast for accessing random bits of data within a stored set but has slower write times and can be more expensive, comparatively. NOR flash is best for storing data like firmware and boot code.

• NAND flash: NAND flash uses “NOT AND” Boolean logic gates, which is more cost-effective for storing larger files. NAND flash excels at sequential data access but is slower than NOR flash for random reads. For these reasons, NAND flash isn’t well suited for direct code execution, but it is an excellent option for long-term and large-scale data storage.

On a technical level, NAND flash memory stores data by using an arrangement of floating-gate transistors that function similarly to NOT AND logic gates. Each NAND memory cell contains both a control logic gate and a floating logic gate, separated by a thin oxide layer.

Through the Fowler-Nordheim tunneling process, binary data can be persistently stored within flash memory by sending an electrical charge into the cell, at a voltage that traps electrons in the floating gate. In turn, data can be erased by removing the charge. 

While flash memory comes in both NOR and NAND varieties, the phrase “flash storage” almost always means NAND flash. Although NAND flash storage media does degrade over time, needing regular hardware replacement after a few years, NAND memory meets many criteria for ideal operational storage in most situations.

The following are some important characteristics:

• Nonvolatile: NAND flash can retain data without the need for continual power. 

• Block-based: NAND flash writes and erases data in large blocks of information, as opposed to a byte-by-byte approach. 

• Efficient at scale: As a result of its block-based structure, NAND flash is well suited for sequential computing tasks that require large amounts of data storage and access at scale. NAND flash is a more efficient and higher-density option compared to NOR flash.  

• Durable: Both NAND and NOR flash memory units are much more durable than HDDs. Flash memory units do not contain moving parts, which can be a point of failure for HDDs. As such, NAND flash is a far superior choice when durability is a factor. NAND flash has become the go-to choice for data storage that might need to be physically transported, and it is consistently used for devices like portable external hard disks drives or laptops. However, while solid-state memory media is definitely more durable than memory units with moving parts, there is a finite limit to how many times a flash memory unit can be rewritten. While nonvolatile flash memory is commonly thought of as indefinitely secure, these types of units do degrade over time, either through overuse or general decay. Best practices dictate that data stored long-term on flash memory units should be regularly validated and periodically duplicated to fresh memory hardware. 

Valuable for its compact size, fast speeds, rugged durability and high storage capacity, flash memory has become a preferred solution for many devices. These devices include memory cards, USB flash drives, smartphones, laptops, digital cameras and similar portable ones. But flash memory is also favored for larger and more physically permanent computing as well. 

Used as the main storage component of any all-flash array (AFA), solid-state drives (SSDs) are typically semiconductor-based storage devices that use NAND flash memory. Each NAND flash memory chip consists of an array of blocks, known as a grid. Each block in the grid is made up of an array of memory cells, referred to as pages or sectors.

Engineers categorize individual memory cells based on the number of bits they can store. Single-Level Cells (SLCs) store 1 bit of information, while Multi-Level Cells (MLC), Triple-Level Cells (TLCs) and Quad-Level Cells (QLC) contain 2, 3 and 4 bits.

Each type of cell comes with its own benefits and challenges. For instance, SLCs are known for high speed, but also high prices, while more affordable QLCs might be less reliable than their more expensive alternatives. Depending on the type of cell, each grid within the SSD can store between 256 KB and 4 MB of data.

In a typical personal computing system, the computer's main central processing unit (CPU) acts as a controller dictating any reading and writing jobs to memory. An AFA can have other hardware or software to help manage these tasks across the entire array of networked SSDs. 

Originally, SSD drives were compatible with the SATA (serial ATA) memory storage interface first developed for older HDDs. Capable of max speeds of around 550–600 MB/s, SATA SSDs are faster than traditional HDDs.

However, SDDs designed for the newer nonvolatile memory express (NVMe) protocols can take advantage of the high-speed PCI express (PCIe) interface. This design allows them to achieve even better performance, with data transfer rates from 3,500 MB/s up to 14,000 MB/s. In other words, NVMe SSDs are about 20 times faster than older SATA SSD technology.

Offering more than increased throughput, NVMe drives also outperform SATA drives on latency tests and boast support for thousands of simultaneous command queues. For these reasons, NVMe SSDs—and AFAs featuring these types of drives—are ideal for demanding tasks like high-performance gaming, content creation and enterprise storage. 

According to Gartner, SSDs are emerging as the go-to storage platform to support structured data workloads, which is fueled by innovation around NAND flash and storage class memory (SCM) technology.

Modern all-flash storage arrays integrate advanced software (and occasionally dedicated hardware) with high-performance NVMe drives to efficiently pool resources from multiple drives.

Going one step further, the best all-flash arrays leverage an even newer protocol, NVMe over Fabrics (NVMe-oF) to maximize data transfer speeds and latencies throughout a SAN. These systems enable hosts to connect and communicate directly to storage by integrating custom NVMe-oF modules in place of an array of separate NVMe SSDs. 

By pooling resources, AFAs create ample storage capacity with necessary redundancies while offering performance and storage management speed-ups. AFAs also use advanced data reduction techniques, including data deduplication, compression, snapshots, replication and thin provisioning, to dramatically improve storage efficiency. 

Most all-flash array vendors also offer sophisticated monitoring, data management and data protection software to empower system administrators responsible for tracking resource usage, system health and other relevant metrics. These tools help organizations avoid bottlenecks and quickly respond to any potential technical issues or incidents.

Beyond management software, AFA products like the Pure Storage FlashBlade, Dell Unity XT or HPE Alletra also offer plug-ins and extensions to allow all-flash storage systems to quickly adapt to changing workloads and seamlessly integrate with hybrid cloud or VMware environments. 

The following are the two main types of architectures used for all-flash storage technology.

Scale-up AFAs follow an older architecture that allows for extra storage to be added to an existing array.

While scale-up architecture allows for a simple and cost-effective way to increase capacity, this method can lead to bottlenecks because the central storage managing controller can become a single point of failure. 

AFAs that use a scale-out architecture allow for a simultaneous addition of more storage capacity and compute power, by adding more nodes to the system.

Scale-out AFAs are a preferred choice for enterprise storage with the potential for rapidly expanding storage needs because this type of architecture ensures consistent performance even as a system grows.  

All-flash arrays (AFAs) offer a wide range of mission-critical benefits, from scalability to security. These benefits are some of the most impactful advantages of an AFA.

As flash memory technology has improved in performance, availability and affordability, all-flash arrays (AFAs) have grown in popularity across many use-cases and industries:

• Virtualization: AFAs have been integral for providing the critical performance and feature support for hypervisor software running virtual machines and virtual desktop infrastructure (VDI). For virtualization, AFAs are used to sustain the high data throughput with minimal access latency necessary for uninterrupted virtualized workloads. 

• Big data and artificial intelligence (AI): AI models are also known for demanding large datasets for training and operation. AFAs are suited for big data applications like running complex AI algorithms that require significant storage performance. 

• Disaster recovery (DR) and backup: Although historically cost-prohibitive, improving flash memory prices and performance have made AFA solutions a viable option for backup and disaster recovery. An AFA’s fast read/write speeds are suited for multistage backups. 

• Financial services: Advances in financial technology (fintech) have ushered in an era of high-frequency trading demanding near-instant data processing speeds. AFAs are a tremendous asset to financial services providers, facilitating instantaneous transactions and data analytics. 

• Media and entertainment: From streaming services to online media marketplaces, AFAs are valuable for media and entertainment companies seeking to store vast quantities of large data files and deliver high-speed data transfers on demand. 

• Healthcare: In the healthcare industry, data access can literally save lives. AFAs offer hospitals and healthcare providers rapid data access that can facilitate timely and effective caregiving. 

• Data centers: As modern life and industry grow increasingly dependent on massive quantities of data, data centers are constantly striving to meet demand. AFAs facilitate fast and reliable data access while minimizing downtime.