Despite the fact that (Hyper Converged Infrastructure) HCI is made to simplify IT infrastructure it has to be implemented just right in order to deliver maximum performance and functionality. Much like any technology, I suppose. Below you can find one of our observations that will hopefully spare you the need to learn this the “hard way”.

You probably know that HCI leverages storage replication between the nodes in order to ensure that your data is available in the event of hardware node failure or planned maintenance, even though one of the leading vendors (Nutanix) has a unique Data Locality feature for read requests, all of them use the network for the writes. Evidently it is very important that the remote writes be acknowledged as fast as possible, as the replication is synchronous and the write operation is not completed until the confirmation is received. Too many broadcast frames can easily overwhelm your HCI nodes and you will not be able to get the normal performance that you would receive if they were the only devices in the layer two domain. In our experience, We’ve seen big enterprises make the simple mistake to use the same layer 2 domain for their HCI nodes and other VMs/devices. While your HCI environment will still operate, chances are you will not get the full performance from your investment. This could be easily avoided if you use separate layer two domain for them. And if you still think that you are on the safe side, because you have e.g. just 50 VMs in the same broadcast domain there are a lot of things that could go wrong with your layer two network and the result is total meltdown. As renowned expert Ivan Pepelnjak is trying to remind us almost every year that “Layer-2 Network Is a Single Failure Domain”, please- keep it simple and safe-dedicate separate layer 2 network for you HCI nodes.