Hyperconverged Infrastructure

Hyperconvergence Is Great for the Remote/Branch Office

Hyperconverged Infrastructure (HCI) vendors report success selling HCI solutions to organizations for Remote Office/Branch Office (ROBO) deployments.  The reasons for this are the same reasons why HCI has proven so popular amongst Small and Medium Businesses (SMBs).

HCI solutions remove much of the complexity involved with storage.  Before HCI, administrators wanting shared storage for clustering or virtualization were forced to choose between shared JBOD, SANs or NASes.

JBODs had to be wired into each system that shared the storage, and usually no more than two systems could share this storage.  In addition, the JBOD chassis itself is a single point of failure, though one could just as easily design a solution with two JBOD chassis, for additional cost.  JBODs are great for ROBO deployment, but only if one is certain they will never need more than two servers’ worth of compute power.

SANs are designed to be resilient against virtually any failure and to have multiple hosts accessing a single SAN.  Unfortunately, they are also expensive, complicated to set up, and annoying to manage.  SANs are economically more efficient where there are a number of hosts accessing the same shared storage, but they are also unwieldy and difficult to manage at scale.

NASes don’t have most of the management concerns of SANs, and they can be found for not much more cost than JBODs.  The downside to NASes is that they aren’t big on resiliency.  While NFS 4.1 has aimed to solve this by bringing multipathing to NASes, adoption has been slow.

In short, before HCI you could have your storage be inexpensive, resilient, or easy to use and scale.  Pick any two.

The HCI Way

HCI aims to provide a solution so simple that it doesn’t require management, is easy to scale, is resilient against almost any kind of failure, and inexpensive enough to be used even by SMBs.  It’s not hard to see why the marketing types love to tout HCI as the solution to all ills.  On paper, it looks great.

In reality, HCI has a whole lot of things in its favour and a handful of gotchas.  From a ROBO standpoint, HCI is useful because it has a low floor cost.  This means that the minimum size deployment isn’t particularly expensive when compared to solutions delivering a similar level of capability.

If you grind your vendors, you might be able to get a two-node JBOD cluster for cheaper than a three-node HCI cluster, which would probably come in about the same as an entry-level two-node NAS.  SANs are typically significantly more expensive than all of those offerings, even before you buy the compute nodes.

HCI is also compact.  Many 3 and 4 node solutions fit in 2U.  Others will deploy in 1U per node, however, no additional space has to be devoted to a separate storage solution.  While this doesn’t matter as much in large datacenters, this can be quite important in smaller offices.

The small footprint means that an HCI cluster, Uninterruptible Power Supply (UPS) and switches can comfortably fit in a small 12U mini-rack, and can likely be cooled by the standard office HVAC system.  Given the capabilities of modern HCI solutions, such a deployment could conceivably run a location of up to 100 people.


HCI’s biggest downside is that most implementations require a large number of network connections per node.  The average is 4, but some solutions will have 5 or even 6 network connections per node.

HCI solutions – like all resilient network attached storage solutions – also expect to have two switches available in order to provide network resiliency.  The net result of these network requirements are that organizations deploying HCI solutions into ROBO environments may need to ensure that those locations have more network ports available than they would otherwise have to provision.

In context, the network port requirement downside of HCI is relatively minor.  Buying switches with a few additional ports doesn’t increase the cost of the solution by much.  Where it can be impactful is when considering growth.

If a location is expected to grow beyond their initial HCI node deployment before the end of life of the equipment then buying switches with more ports than is required by the initial HCI deployment is advised.  While these network port requirements can be burdensome at extreme scale, they haven’t proven to be a big deal for ROBO deployments.