What is disaggregated hyperconverged infrastructure (dHCI)?

Hewlett Packard Enterprise (HPE) introduced the term disaggregated HCI in 2019 alongside its HPE Nimble Storage dHCI infrastructure platform, although NetApp reportedly came up with a similar framework as early as 2017. In the ensuing years, other vendors have increasingly borrowed and incorporated dHCI concepts into their own infrastructure solutions.

dHCI (sometimes called HCI 2.0) builds off its predecessor, HCI, which gained popularity in the early 2010s. HCI’s primary innovation is aggregating storage, compute and networking into self-contained nodes and providing streamlined oversight and configuration through a software-defined management layer. HCI also enables simplified scaling; teams can add new nodes as needed without the compatibility risks and manual provisioning of earlier approaches.

However, because network, compute and storage resources are bundled together in HCI, teams cannot scale one resource without also scaling the others. Adding 20 TB of storage, for example, might also entail introducing additional central processing unit (CPU) cores. This limitation can lead to over-provisioning, where organizations pay for resources that sit idle.

dHCI aims to preserve HCI’s shared management layer but separates compute and storage into separate pools, giving teams more fine-grained control over resource scaling compared to HCI. While dHCI introduces additional configuration complexity, it can be well suited for unpredictable or intensive enterprise workloads, such as big data and predictive analytics, artificial intelligence (AI) and mixed workloads, where demand for different resources might grow unevenly.

Many dHCI vendors offer hybrid cloud integration, enabling organizations to blend the stability and control of on-premises infrastructure with the flexibility of cloud-based management. For example, HPE Alletra dHCI can connect to the cloud-based management platform HPE Greenlake, giving teams greater operational flexibility. Many dHCI solutions also provide built-in data protection, data reduction, data efficiency, data availability and resiliency services.

Because dHCI is a relatively new technology, few reports have looked at the framework’s prevalence. It likely makes up a small but growing share of the wider HCI market, which is expected to grow to USD 70.50 billion in 2034 at a CAGR of 17.6%, according to Fortune Business Insights. That growth is driven in part by a rise in data center digital transformation efforts and growing demand for robust data security and data recovery capabilities at the infrastructure layer.

dHCI shares a similar set of hardware and software components as standard HCI, with some key differences. Components include:

Storage arrays typically contain solid-state drives or hard disk drives (or both), controllers, network ports and cache memory. They can also provide a suite of data handling capabilities, such as compression (reducing the size of data), snapshots, replication, data protection and telemetry collection. Storage arrays communicate with a separate compute layer over a network, unlike standard HCI, where storage and compute are housed in the same node.

The compute layer is typically made up of multiple servers, each containing CPUs (or graphics processing units - GPUs), memory (RAM) and hypervisors, which run virtual machines (VMs), or digital representations of hardware components. The compute layer is responsible for running workloads but might also contain features related to application performance and efficiency. For example, HPE’s dHCI solution uses HPE Proliant servers, which provide built-in automation tools, predictive analytics and monitoring capabilities.

The management plane sits above the storage and compute layers and is accessible through a network or web client. The management layer provides unified oversight, governance, provisioning and orchestration, among other capabilities. This architecture differs from traditional frameworks, where teams configure and manage compute and storage resources across separate, disconnected planes.

While dHCI can be an effective on-premises infrastructure approach, it might not be appropriate for every use case. Here’s how it compares to other infrastructure frameworks.

In traditional frameworks (also known as three-tier architecture), organizations purchase and upgrade components individually and are responsible for integrating these new components into existing systems. For example, if an enterprise runs out of storage but has sufficient compute capacity, it can buy more drives without also purchasing upgraded CPUs and RAM. Organizations can also mix and match components, choosing one vendor for SSDs, another for servers and yet another for RAM.

Often, organizations connect physical hardware with VMs through a software layer called a hypervisor, enabling a single server to effectively act as multiple independent computers. Enterprises can use various approaches, including network attached storage (NAS), storage area networks (SANs) or direct attached storage (DAS), to store VM data. They can also choose from various hypervisors, including VMware ESXi, Microsoft Hyper-V or Linux’s kernel-based VMs, which each present different features and integration options.

One downside with this approach is that older components might be unable to use newer components’ distinct features or advantages, introducing bottlenecks or other issues. For example, a company might upgrade from SATA SSDs to NVMe SSDs, which generally offer superior throughput. But unless the organization has also upgraded legacy storage controllers and connectors, it cannot take advantage of NVMe’s performance benefits.

Finally, governance and management can become disjointed and fragmented, as teams manage compute, networking and storage individually, rather than through a unified and abstracted software layer.

Converged infrastructure (CI) offers compute, storage and network resources as a bundle, addressing some of the shortcomings of traditional infrastructure. With CI, there’s little risk of performance mismatches. The vendor guarantees that bundled components are compatible with each other (although compatibility issues or performance mismatches can arise over time, as organizations update or add components). Technical support is also simplified, as the client organization must interact only with a single vendor for troubleshooting and assistance.

While components are packaged together, they are still treated as individual entities and managed across separate layers. This approach preserves the scalability of traditional infrastructure. For example, teams can add more servers without adding storage or network capacity. However, as with traditional deployments, management can present a challenge. Although the vendor validates the initial configuration, updating a component might require IT teams to reconfigure existing components, or might introduce bottlenecks as older components struggle to keep up.

With HCI, vendors not only bundle components together but also present them through standardized, fixed nodes. This strategy simplifies deployments: To expand capacity, organizations can purchase more nodes. While traditional infrastructure and CI are largely hardware-defined, HCI places a greater emphasis on software. Monitoring and configuration take place through a single management plane, rather than through separate network, compute and storage frameworks.

HCI’s simplified configuration makes it a good fit for relatively predictable workloads, especially when demand for compute and storage scale at about the same rate. It’s often used to support virtual desktop infrastructure (VDI), where multiple virtual computers are hosted on a cluster of servers, rather than tied to physical devices. As the user base expands, the organization can provide more storage and compute capacity by adding nodes.

A newer feature developed by VMWare called HCI Mesh bundles compute and storage into nodes, then compiles those nodes into larger clusters. Instead of remaining siloed, these clusters can access each other’s storage resources. This strategy can create a more dynamic, flexible environment by enabling clusters to share storage capacity, instead of leaving excess storage unused.

While traditional HCI exclusively supported commodity hardware (which uses commonly available, interchangeable components, such as x86 servers), many modern solutions now provide integration with specialized components such as GPUs and storage class memory (a high-speed memory storage technology that combines elements of RAM and NAND flash storage).

One limitation is that storage, compute and networking are composed into a shared layer, limiting customization and scaling granularity—and potentially introducing inefficiencies. Adding 10 TB of storage, for example, necessarily entails adding more compute, as these components are coupled.

Finally, although HCI uses off-the-shelf hardware, its software components—including management, automation and virtualization features—are typically delivered as a tightly integrated package. This software-defined approach can contribute to vendor lock-in and higher switching costs.

dHCI shares HCI’s streamlined, unified management (IT teams can manage and configure resources through a single control plane) but keeps servers and storage arrays separate to enable independent scaling. This approach can improve operational efficiency and lower total cost of ownership (TCO), as organizations are less likely to waste funds on unused resources.

If an organization’s storage and compute needs are unbalanced, dHCI can be a good option. For example, AI workloads might require high ongoing GPU usage but only modest storage capacity. Alternatively, a data backup or logging workload might use multiple high-capacity drives but relatively little compute.

However, for workloads that use an equivalent amount of storage and compute, standard HCI is often sufficient and less operationally complex. Also, dHCI generally requires more IT resources because, although it provides a centralized layer for high-level management, some compute and storage management must still be handled separately through individual control planes.

Also, as with HCI, dHCI vendors tend to package resources together. Even though teams can scale compute and storage individually, they must typically do so through the same vendor. This limitation can lead to vendor lock-in and can make it more costly and difficult to switch to a new infrastructure strategy in the future.

Cloud-based infrastructure solutions such as infrastructure as a solution (IaaS) can provide some of the same benefits as dHCI, including streamlined management, fine-grained scalability and disaggregated, modular resource blocks. However, with IaaS, the vendor maintains ownership of resources and uses a pay-as-you-go model to charge clients based on usage.

In dHCI deployments, meanwhile, organizations purchase hardware and software and maintain full ownership of these resources on premises. This approach is often preferred for intensive workloads, where cloud-based usage might be too expensive at scale, and where local or edge provisioning can deliver more predictable performance.

dHCI might also be appropriate for firms in highly regulated industries, where compliance and auditability require a greater degree of control over data and infrastructure resources. However, this approach necessarily uses more IT resources, as infrastructure is managed internally rather than by a third-party vendor.