HCI Benefits Without HCI, Part 1: Physical

Do you need to buy an HCI product to get the benefits of a hyperconverged system? I don’t believe that storage and compute need to be on the same physical server to get the big HCI benefits. I think you can get some of the biggest value that HCI delivers without using HCI. The big HCI benefits I see are:

  1. Modular, scale-out expansion
  2. No dedicated storage network, no Fibre Channel
  3. Simplified management

In this article, I’ll look at the first two, which are primarily physical. Simplified management will get its own article. A defining characteristic of HCI is using local storage inside hypervisor nodes as shared VM storage. If you choose not to use HCI, then you are going to have a separate storage system for your VMs. I would also like to be clear that I am not saying HCI is a bad idea, just that there may be other ways to get its greatest value.

Modular, Scale-Out Expansion

An HCI solution is built from a series of servers with local storage. The same servers create a storage cluster and a hypervisor cluster. With HCI, the storage and hypervisor clusters overlap, using the same physical hardware. When the workload on your HCI grows, you simply buy nodes to expand both the hypervisor and storage clusters. You only need to plan for a few months of growth, then buy the additional capacity as you need it. This short-term forecasting and planning is more reliable than guessing what will happen over the five-year lifespan of a scale-up storage array. There is a challenge when either compute or storage capacity needs to grow while the other does not. With HCI, you must often buy both compute and storage together, even when you only need one or the other. We do see HCI solutions with different node configurations that are either storage or compute heavy to accommodate different workloads. Some HCI solutions support compute-only nodes; others have storage-only nodes. All this variation suggests that customers want a lot of flexibility. Combining storage and hypervisor in one piece of hardware may not always be the best solution.

To get the same benefit without HCI, we need a scale-out storage cluster and a separate scale-out hypervisor cluster. The hypervisor cluster is easy: just use a bunch of compute-heavy servers without local storage. When you require more compute capacity, simply buy more hypervisor hosts. Scaling out storage is also possible; products like SolidFire use a scale-out architecture. When you need additional storage capacity or performance, you add nodes to the storage cluster. Naturally, the storage cluster will be built from a set of x86 servers, which have a cost. This storage system cost will be offset by not needing the hypervisor cluster to provide compute resources to run the storage cluster. The hypervisor cluster will not lose 10–20% of its compute capacity to run the storage cluster. Separating the storage and compute allows each to be scaled independently, possibly leading to lower cost. A modular, scale-out architecture does not have to be HCI.

No Dedicated Storage Network, No Fibre Channel

HCI solutions eliminate the arcane Fibre Channel network, along with the specialist skills to configure zoning and selective storage presentation. HCI uses Ethernet for storage. The same physical network is used for VM management and VM guest networking. Storage networking is converged. Ethernet is a perfectly sound network; it was used for storage long before HCI came to the market. vSphere supports iSCSI and NFS for datastores. Hyper-V can use iSCSI or SMB shares. Converging the storage network is perfectly routine for non-hyperconverged environments. Eliminating the dedicated storage network does not require HCI.

Sometimes HCI Is Still Right

Even though you can get some of the HCI benefits without HCI, there are still places where HCI is going to be a better solution. The obvious case is when the extra servers for a scale-out storage system are a big deal. Every HCI solution can be deployed as a three-node configuration. Many can do two-node configurations or even sites that have a single node. Most scale-out storage solutions will need at least three servers to get started, along with at least two hypervisor hosts. Separate storage and hypervisor clusters aren’t going to beat HCI for small sites.

The hyperconverged trend has delivered some significant benefits to enterprise IT. A few of those benefits are driven by the physical implementation. For some uses, it is possible to get the same benefits without using a hyperconverged product. As you are evaluating vendors for new infrastructure, it is important to consider whether what they offer is truly unique to their product.


Share this Article:

The following two tabs change content below.
Alastair Cooke
Alastair Cooke is an independent analyst and consultant working with virtualization and datacenter technologies. Alastair spent eight years delivering training for HP and VMware as well as providing implementation services for their technologies. Alastair is able to create a storied communication that helps partners and customers understand complex technologies. Alastair is known in the VMware community for contributions to the vBrownBag podcast and for the AutoLab, which automates the deployment of a nested vSphere training lab.
Alastair Cooke

Latest posts by Alastair Cooke (see all)

Related Posts:

Leave a Reply

Be the First to Comment!