Storage Predictions for 2024 and Beyond (Part II - Systems)

This is the second of a multi-part series looking at technology infrastructure predictions for 2024 and beyond. It builds on discussions from previous posts to predict what we might expect in the segment of storage systems from a short and long-term trend perspective.

Storage Predictions for 2024 and Beyond (Part I – Media)

Background

One of the most enduring technologies of the last 70 years of computing has been the persistent storage array, also called a storage appliance or storage system. EMC (acquired by Dell in 2015) is credited with building Symmetrix, the first ICDA (integrated cached disk array), in 1990. This platform built on commodity drives (the 5.25” models of modern HDDs), used RAID technologies and cache memory to accelerate I/O. The ICDA added resilience, performance, and scalability to monolithic storage subsystems from the mainframe era.

Evolution

From the 1990s onwards, all major infrastructure vendors transitioned towards the ICDA model, initially supporting SCSI connectivity, then moving onto Fibre Channel and Ethernet protocols such as iSCSI. In the early 1990s, NetApp pioneered the NAS filer, introducing file-based protocols to storage systems.

Hitachi Data Systems (now Hitachi Vantara), NetApp, IBM, HPE, EMC and Dell (first independently, then with EMC) have all created and developed scale-out and scale-up storage systems that have rapidly increased in capacity and performance. The availability of reliable commodity components has enabled the storage array to transform from a bespoke 42U rack of equipment to a standardised 1U, 2U or perhaps 4U rack-mounted appliance.

To understand just how far the industry has evolved, at the turn of the new millennium, a typical 42U storage array from Hitachi weighed 1700kg and consumed 14kW of three-phase power. It delivered the capacity and performance of a single modern SSD, such as the Solidigm D5-P5430, which is capable of 7GB/s read bandwidth (3GB/s write) and stores 30TB of data in a device weighing less than 100g. Power consumption is a paltry 25W.

How far we’ve come.

Standardisation

Look across the market today, and all storage appliances have settled on either a scale-up or scale-out architecture. Increasingly, scale-up designs are more popular, as the capacity and performance of individual controllers exceed the needs of most applications, especially when considering the “blast radius” or impact of an appliance failure. Features that were once differentiators are now table-stakes capabilities, such as high scalability, high availability, resiliency, platform efficiency, operational flexibility, and connectivity.

We expect systems to run 24/7 with close to or 100% uptime, offer online maintenance, deliver consistent performance and be autonomous in operations. The days of needing teams of storage administrators are over.

New Differentiators

If all the existing platforms look similar, how do we differentiate between them? Inevitably, new requirements and considerations come to bear as computing environments evolve. Modern requirements include the following.

Cyber-resiliency – the ability to act as the last line of defence for ransomware attacks and to enable fast recovery in the event of a system breach.
Fleet-management – the capability to manage dozens of individual systems through a portal, typically delivered as a SaaS application. Fleet management solutions should provide significant operational “value-add” that reduces the operational burden of system and storage administrators.
Flexible consumption – the choice to mix and match components to meet demand from the business efficiently. The days of unused hardware sitting idle are over and don’t need to be part of modern infrastructure deployment.
Financial flexibility – the choice to pay for infrastructure as a capital purchase, a lease or through consumption models. Modern financial flexibility should also include the capability to move hardware between platforms without a penalty.

Each of these requirements echoes the capabilities of the public cloud, where storage is available on-demand. While it is challenging to deliver fully on-demand infrastructure in an on-premises environment, vendors have the power to address the operational aspects of persistent storage while using architectural designs and standardisation to optimise the footprint of deployed infrastructure.

Predictions

With increasing similarity between vendors, what can we expect from the future of on-premises storage appliances?

Declining costs and increased reliability. This prediction is not that challenging to make but does have some nuances. In 2024, NAND prices are set to rise, while the unit capacity of the largest SSDs hasn’t exceeded 30TB for more than half a decade. Prices will continue to fall, while reliability will increase, simply due to the improvements in media technology. However, continuing this trend may be more challenging for some vendors than others. The storage appliance market has typically been a cut-throat business, with narrow margins and lots of competition.

The end of enterprise arrays. The monolithic systems we discussed from the turn of the millennium are all but gone. Vendors can build systems from off-the-shelf components without the need for single-rack designs and three-phase power supplies. However, we should highlight the term “enterprise”, which generally means high-end resilient technologies built for mission-critical applications. Modern enterprise storage has merged with mid-range systems to the extent that most applications can be delivered through the mid-range footprint. Within a few years, we expect the terms “enterprise” and “midrange” to be obsolete.

Transition to all-flash. Some people may say the all-flash data centre has been a long time coming. However, most storage appliance vendors are already singly focused on all-flash systems, with some no longer (or never) offering hybrid or HDD systems. With each generation, the hard drive introduces further compromises in performance in order to gain additional growth in capacity. This makes these devices less suitable for active workloads and to be pushed further to the archive tier. All-flash is the default deployment model for the majority of enterprise customers and will gradually phase out HDD and hybrid systems over the next five years.

Feature stagnation. What else can we add to storage appliances? The most recent innovations have focused on flexible consumption, cyber-resiliency, and fleet management. These are not platform features but changes to the operational model. We think there will be little or no new feature innovation from vendors that affect the direct storage and retrieval of data from storage systems. Instead, feature improvements will all be about operational and cost efficiencies.

Predictable upgrades. As systems have migrated to commodity hardware, so we’ve seen the introduction of upgrades aligned to new processor architectures. Intel Sapphire Rapids introduces PCI Express 5.0 and DDR5 memory, while PCIe 6.0 and 7.0 are on the horizon. CXL will also be available for performance or operational enhancements, such as increased memory footprint (or persistent memory for metadata/cache).

Niche opportunities for alternative architectures. While the Intel x86 ISA has dominated storage infrastructure, we’ve seen Arm offer an alternative route for systems design (with SoftIron), possibly with the use of SmartNICs. GRAID has used a GPU to create a “distributed RAID card”, while Nebulon has progressed with distributed storage based on a custom SmartNIC. Nyriad has developed a storage appliance using GPUs that keeps the HDD alive. Weka has announced a tech preview using NVIDIA BlueField DPUs, where the WEKA client runs on the DPU rather than on the host. At present, these solutions have either been deployed as proofs-of-concept or are still waiting for more widespread adoption.

Platform Integration. We highlighted the integration aspects previously, referencing NVIDIA GPU Direct as an example. As AI drives more conversations, we can see further integration developing, with closer operation between storage and compute. This could be especially the case with containerised applications and further use of SmartNICs. We think this is a nascent area of evolution that could be a source of future innovation.

Service-based pricing. We’ve already highlighted the requirement for financial flexibility, which includes consumption-based pricing. Legacy vendors now offer flexible consumption models, but these appear to be disguised leasing agreements and not true capacity on demand. In a competitive market, the $/GB ratio still applies despite the differences in features and functionality. Vendors such as Pure Storage have introduced new capabilities, such as power and rack-space rebates, in an attempt to put clear blue water between the leasing-style pricing and real consumption models. We expect to see further evolution of these schemas, perhaps to include sustainability and emissions rebates, as vendors look for increased differentiation.

Rise of SoHo Devices. We highlighted this trend last year, echoing the Innovator’s Dilemma scenario, where smaller, nimbler entrants eventually challenge the incumbents. The storage industry has many second-tier “good enough” vendors, although some have struggled to gain traction and gone out of business (Violin Systems, Pavilion Data, StorCentric are examples). A decade or two decades ago, these tier-2 companies would have been quickly acquired, but today they operate as independent businesses. Many don’t follow the appliance model, as software-as-a-service has become more mainstream.

The Architect’s View®

The storage market is a sticky business (perhaps not as sticky as data protection). Businesses like predictability and reliability for data assets that are increasingly valuable to staying competitive. Changing storage providers is generally a major decision point, not taken lightly. As a result, the storage market moves relatively slowly.

However, we can see the impacts of our predictions in the relative success and failure of the most prominent players in the market, including HPE, IBM, Dell, NetApp, Pure Storage, Infinidat and others. The evolving dynamics of the industry have enabled Pure Storage and Infinidat (for example) to see constant gains in market share while others remain stagnant or decline. You can learn more about these vendors and their solutions in our series of X-Ray eBooks (available here).

Storage appliances will be needed as long as there is on-premises IT infrastructure. The public cloud has transformed expectations towards a service-based model, which is still in the early days of adoption. The winners and losers in this market will be determined by choice, flexibility, and efficiency. These are all markers of operational improvements driven by, rather than dictated by, new technology. The storage systems market continues to see interesting times.