alphaspirit - Fotolia

Is customizing a hyper-converged platform a good idea?

How much flexibility do you have with hyper-converged systems when it comes to data storage? Discover how to deal with changing hyper-converged resource requirements.

Most hyper-converged products have limited customizability when it comes to the storage they include. One of the basic characteristics of a hyper-converged platform is that it uses a scale-out, building block approach. A cluster is built from a collection of nodes that are all indistinguishable and interchangeable. Hyper-converged products also allow the cluster to grow over time as your needs increase.

So what happens if the mix of resource requirements changes over time? It's common to start with modest data storage requirements compared to compute requirements, but over time the need for storage capacity often grows faster than the need for compute. Do you keep buying more of the same node type and end up paying for way too much compute? It depends on the hyper-converged infrastructure you choose.

A majority of hyper-converged products use two types of storage in each node:

  • Small and fast for performance, usually solid-state flash.
  • Cheap and deep for capacity, usually hard disk drives.
Hyper-convergence is based on the idea that the hardware is capable of all the performance you need. If you need to fine-tune every aspect, it may not be right for your workload.

Some node types are all-flash storage, which can include expensive fast flash and less-expensive high-capacity flash. The faster tier is about performance. There needs to be enough capacity in the fast tier for the data your virtual machines (VMs) access frequently. The cheaper tier is focused on capacity.

Usually, more than 90% of VM data is not accessed in a given day, so it makes the most sense for it to reside on the cheaper tier. Balancing the placement of data between the different tiers is part of the magic in the hyper-converged infrastructure software -- not needing to spend effort on optimizing storage performance is a value proposition for hyper-converged.

The main hyper-converged platform vendors sell fixed-configuration nodes. Each model has a fixed amount of fast storage and deep storage, and you choose the model that is large enough to accommodate your workload. These nodes provide advanced hyper-converged platform features like integrated backups and replication for disaster recovery, but using these features increases your hyper-converged infrastructure storage capacity requirements.

There is no way to upgrade the storage within an existing node; a user must buy new nodes for capacity and performance. Nutanix is notable for having the widest range of models and allowing mix-and-match models within a cluster. EVO:RAIL, Scale Computing and SimpliVity all require clusters be built from the same node type, and have a much shorter list of models as a result.

Unusual workload may require customization

If you have an unusual workload, you may need to fully customize the hardware for your platform. You may need to use a hyper-converged platform from a software-only vendor so you can select a storage configuration that exactly matches your requirements. For example, if you need a larger performance tier and a smaller capacity tier, providing your own hardware will enable almost any blend of parts. But fine-tuning your hardware to exactly match the workload is not the normal hyper-converged method. If you plan to manually fine-tune a hardware purchase for your hyper-converged platform, you may have missed the value of the technology. Hyper-convergence is based on the idea that the hardware is capable of all the performance you need. If you need to fine-tune every aspect, it may not be right for your workload.

Next Steps

The early days of hyper-converged storage platforms

How the different hyper-converged tiers work

Understanding the hyper-converged market

Dig Deeper on Hyper-Converged Vendors and Products