IT administrators have a decision to make when it comes to their hyper-converged infrastructure deployments: Do they deploy HCI as a software layer or purchase HCI nodes as a prepackaged hardware appliance?
Choose between HCI hardware or HCI software
With a continued move to a more commodity-based hardware approach to underpin cloud-based systems, it might appear that hardware-based HCI has had its day. However, hardware-based HCI still has its place for workloads where such tight integration and high optimizations of areas such as storage-to-CPU networking provides the high levels of data throughput that are required for real-time data analytics. Another prime area for hardware-based HCI is VDI deployments, where the complex interactions between CPU, storage and network loads are better served by a hardware optimized environment -- rather than depending purely on software.
Systems that are HCI out-of-the-box include Dell's VxRail systems -- running VMware software -- and the HPE SimpliVity platform. Such hardware systems also tend to have built-in high availability provided using redundant or multiple-redundant hardware components.
Although hardware-based HCI can offer the ultimate in underlying performance, such systems tend to be expensive to procure and to expand, often requiring blocks of a mix of server, storage and networking to be bought where maybe only new storage is required. This need to buy proprietary blocks of resources also means that hardware-based HCI carries a strong risk of vendor-lock in, and can also lead to problems where the vendor ceases to fully support the system or to provide the required hardware blocks to expand the system. They can also be highly power-intensive, and not all data centers can have the power distribution capabilities to support such a system.
There is a different option, though. VMware is the largest player in the software-based HCI market with a suite of software -- driven by vSAN -- providing an HCI-like platform that can be spread across less costly underlying hardware components, with less need for tight coupling of server, storage and networking components. Nutanix continues to perform well as a second-placed vendor -- by value of shipments -- with its software suite, having successfully made the move away from a hardware play some time back. Microsoft also offers an on-premises version of Azure under the Azure Stack portfolio.
Other examples include OpenStack, which has its own HCI-style capabilities, and Red Hat, which provides a system using OpenStack alongside Ansible.
The pros here are that such software can be layered on top of pretty much any hardware, and that the software itself can be cheap -- or even "free" under open source agreements. The negatives are that initial architecting, setup and provisioning are far more complex, and that areas such as monitoring, reporting and management are often needed to be applied as external systems layered on top of the HCI software.
Select from 3 HCI deployment strategies
Having two such different means of implementing an HCI environment then leads to three different approaches that must be considered:
- A complete replacement strategy. Here, all existing systems are retired with new hardware and software being brought in to create a new platform. For most organizations, this won't be financially viable.
- Transitional deployment. Any new software deployments where more tightly integrated resources are preferable are deployed on an HCI environment. As time goes on, old workloads are migrated on to new HCI platforms.
- Workload-by-workload deployment. Consideration is given to the specific needs of each workload with the deployment decision being made on whether tightly coupled resources or a more general virtualized resource platform is right for them. This model will be best for most organizations, enabling optimization of the underlying costs around hardware and software requirements.
Plan for common HCI deployment challenges
However, there are a few things that must be kept in mind when considering HCI:
- The continued move to the use of public cloud and SaaS systems can mean that the architecting and deployment of an on-premises HCI platform might end up having something of a limited life as the performance of public clouds continues to improve. The resource flexibility of public clouds already far exceeds what most private clouds can provide, and availability figures at the platform level are exceedingly high.
- Any HCI environment must operate as a peer against the rest of the environment. Not only must the HCI environment be a holistic part of the overall platform, but it should be monitorable and manageable through the same pane of glass that the rest of the platform is managed through. Avoid systems that have proprietary management software in place that require their own sysadmins.
- Workload portability is a must. Many HCI platforms are used for VDI images or containerized software instances. Check that such images are not tuned to make use of the specific capabilities of the underlying HCI environment and that the workloads can be easily migrated to an alternative platform if required.
HCI is still a suitable way of supporting specific workload types, such as VDI and real-time data analytics. However, it's a case of buyer beware. Make sure that you go into any deployment decision with the right reasons for using it.