Admittedly late to the market, NetApp HCI was announced in mid-2017 and started shipping a few months later. Most analysts were quick to pounce on the solution and attack its positioning as a hyperconverged solution. Pushing all of that criticism aside, what NetApp released is a nice package of compute and storage in a small footprint, 2 rack unit chassis.
With NetApp HCI, there is no compromise in the storage department, utilizing the full Solidfire Element OS with dedicated CPU’s running the storage. Versus a stand-alone Solidfire node, the NetApp HCI storage nodes offer only 6 drives per node (stand-alone nodes have 10). A similar limitation exists with Flexpod SF storage nodes, based on the Cisco C220 M4 with only 8 small form factor drives. In both cases, Element OS simply uses the number of drives available.
Perfect for greenfield deployments because of the NetApp Deployment Engine (NDE). A lot of the value provided by NetApp HCI is around their deployment workflow. It is impressive to see a full vSphere deployment in under 30 minutes. That time does not include all the time needed for wiring and planning. That time could be several hours prior to deployment depending on the size of the deployment.
The NetApp Deployment Engine is easy and straight forward web interface that accepts all of the data from the planning document. With all the IP’s and connections pre-planned in an Excel document, you simply plug in this data into NDE and it takes and builds the compute and storage nodes, installs ESXi, deploys a vCenter (optionally), and then brings it all together with some pre-built LUNs on the Solidfire. It also deploys the Solidfire plugins for vCenter onto the newly deployed vCenter Server Appliance.
While you have the normal Solidfire web interface, everything you need can be accomplished inside of vCenter. That’s no different than a discrete Solidfire cluster with the vCenter plugin, but it is nice that it is an integrated deployment.
I was surprised by the amount of cabling required – 2 x 10Gbps ports per storage node and 4 x 10Gbps per compute node plus several 1Gbps IPMI ports and 1Gbps out-of-band management on the compute and storage. NetApp has heard the complaint and plans to address it. Later versions of NDE will offer a flexible deployment where all the ports are not necessary, but the first iterations of NDE require all ports to be connected.
My one wish was that NetApp somehow packaged a shared network in the chassis of each node, allowing easy interconnection of chassis from one to another and minimizing the number of ports required to cable this thing. You certainly can’t achieve that using an OEM compute vendor’s chassis as the basis. It would take a lot more research and development. But it would greatly improve the overall install and serviceability of the solution.
Room for Growth
This is NetApp’s first foray into compute hardware. While the company has used OEM compute behind Solidfire nodes, it has never offered its own compute nodes.
Compared with other converged infrastructure solutions, NetApp HCI needs to do more to ease long term operations. NDE is a great model for Day 1, but it doesn’t do anything to help beyond Day 1. For continued operation, you use the NetApp Solidfire plugins for vCenter and
NetApp should be qualifying the drivers and configuration of the system as a whole – so compute layer with vSphere and all of its Solidfire nodes together. The testing should be a standard practice, but for customers not using NDE, there should be some packaged intelligence of what software inter-operates well together.
NetApp HCI: Is it hyperconverged or is it not? Honestly, does it matter? My take is NetApp HCI is much more of a converged infrastructure than a hyperconverged solution. I get the point that its all technically in the same 2 rack unit chassis, but it does not colocate the storage and compute in the same physical node. And folks at NetApp tell me that is intentional. They did not want to degrade the powerhouse storage platform that Solidfire offers by stealing compute away from compute.
NetApp is the first to tell you they’ve solve independent scale of compute and storage in the hyperconverged space. I get that also in theory, but it is no different than how we have solved those problems without hyperconverged solutions.