The Future of Data Storage: Emerging Trends in 2024

2024 is coming and we asked our VP of Product Strategy, Sergei Platonov , to share his thoughts on upcoming trends in storage industry in a short article. Do you agree with Sergey or you see other trends that will lead the market? Let’s discuss in comments!

New Generations of NVMe Storage

The demand for storage system performance is driving hardware updates. Currently, many servers support PCIe Gen 5 and CXL. Leading storage manufacturers have released new NVMe models capable of nearly 3 million 4k IOps for reading and 14GBps. More models from almost all vendors are expected in 2024. However, ensuring high availability with such high-performance solutions remains a challenge. Storage cost and density optimization are also becoming crucial in technology development. ?

There are two main parallel threads in technology development: 3D VNAND (and similar technologies) and logical scaling (QLC, PLC, etc.).

Samsung is expected to become a leader in the first approach, planning to release 300-layer memory in 2024. SK Hynix discussed its 8th generation 4D NAND with 321 layers in spring 2023, but its release is not expected before the first half of 2025. ?

QLC memory still occupies a very small percentage in data centers and large companies but is gradually finding its place, especially for AI-related tasks characterized by high-intensity reading. Both methods of increasing storage density will help reduce operational costs for using large volumes of flash memory. For mixed workloads, QLC's application without Optane is challenging; the current SCM offerings based on SLC NAND do not provide the necessary performance level, making the solution overall economically unfeasible. No new developments in this market are expected in 2024-2025. ?

Rising Interest in CXL

In 2023, there was significant interest in CXL, with many devices such as memory expansion modules, memory-semantic SSDs, servers, and switches being introduced. Software manufacturers for composable data centers also supported the technology. While these solutions bridge the gap between local memory and high-performance storage systems, there are concerns that only a few companies will be able to exploit the technology effectively. Many have not yet mastered dealing with two Numa nodes, and adding more "layers to the cake" with very different delay characteristics could complicate matters. Many devices supporting CXL 2.0 are expected in 2024, with numerous announcements for CXL 3.0 solutions. Significant changes in applied software in terms of support and real-world cases are anticipated; otherwise, the technology might quickly become obsolete. ?

Composability Challenges

The personally favored composable approach to storage systems is somewhat "stalled" due to 1) the relatively low prevalence of networks supporting RDMA and 2) the lack of conventional software for storage management and services. ?

The first problem is beginning to be addressed with the transition to NVMe/TCP and the use of specialized network devices that take on initiator functions and emulate local NVMe device. Software is being developed by various vendors, but there is no single "gold standard" yet. Composability is seen as an excellent means for use in private and public clouds focused on AI services. The type of storage system workload (and its variability over time) in these environments is not entirely evident even to specialists, so maximally flexible and efficient solutions are required. ?

Computational Storage

Another technology not yet finding the expected market response but believed to have a future is Computational Storage.

At FMS this year, Kioxia discussed one application scenario - RAID offload. Although computations do not occupy much CPU resource in our products, offloading them to the storage level could help relieve the network in disaggregated environments and memory.

The discussion is now about hundreds of GBps per server. The connection of accelerators was recently added to OpenZFS. We are closely monitoring this technology and believe that Computational Storage will increase its presence in Data Intensive environments. ?

Special Attention to DPUs (Data Processing Units)

It is worth discussing DPUs (Data Processing Units) separately. In the context of the development of CCX, CCIX, as well as the increasing adoption of Computational Storage and Composable Storage, we can expect a significant increase in interest in DPUs, especially in HPC (High-Performance Computing) and AI (Artificial Intelligence) environments. DPUs are specialized processors designed for data processing, which significantly offload central processors and enhance overall system efficiency. They are becoming key components in modern computing environments, where high-speed processing of large volumes of data and resource optimization are required.

Unstructured Data

With the growing interest in AI, the expenses for storing unstructured data are set to rise rapidly. Despite the potential hype, the demand for both existing market solutions and the development of proprietary solutions by major cloud providers for storing unstructured data will increase. There will also be a growing interest in tools that reduce data ownership costs and provide data lifecycle management. ?

Mergers, Splits, and Disappearances

In 2024, we can expect a wave of significant mergers, splits, and possibly the disappearance of some companies or at least several product lines. These events will undoubtedly impact the industry's state, with a specific focus on the technological aspects.