WavebreakMediaMicro - Fotolia
NetApp's approach to hyper-converged infrastructure isn't the same as other HCI vendors', but it can still work, especially for large data centers.
NetApp packages compute and all-flash storage into a single hyper-converged infrastructure (HCI) appliance with its SolidFire Element OS. NetApp HCI provides the same simple building block scaling model that attracts customers to HCI, but companies don't have to add storage when all they need is more compute -- and vice versa. Yet, NetApp HCI doesn't actually meet the true definition of hyper-converged infrastructure.
The NetApp HCI appliance uses the same high-density, four-server and two-rack unit appliance that Nutanix and other hyper-converged infrastructure vendors have used for years. But rather than running its storage management processes under the hypervisor in a virtual machine (VM), as Nutanix does, or as a kernel process, a la VMware's vSAN, NetApp dedicates some nodes to running the SolidFire Element OS and others to running VMs under VMware's vSphere.
Industry experts don't agree on the origin of the term hyper-converged, but there is general concensus that it refers to systems that are more integrated than Vblocks, FlexPods and similar converged products that bundle storage and servers into a single SKU.
Storage analyst Arun Taneja and former VCE CTO Steve Chambers seem to have coined the term hyper-converged independently. Knowing both of them, I don't buy the argument that the hyper in hyper-converged comes from hypervisor. Hyper-converged must mean an architecture where storage management and user applications share the same CPU. Without this distinction, vendor marketing deparments can control the definition of hyper-convergence.
NetApp HCI is -- and isn't -- hyper-converged
Technically, the NetApp HCI appliance is a disaggregated software-defined architecture with dedicated storage and compute nodes in a single appliance. Strictly speaking, it's not hyper-converged, but companies can deploy it to address the same use cases as HCI appliances, with significant advantages.
First, NetApp HCI decouples compute and storage as a cluster scales. Most HCI vendors sell nodes with varying amounts of storage and compute, but still require customers to add an entire new appliance when they need to scale up. So shops must add some storage when they want to add compute, and they have to add some compute when they only need more storage. If the storage process runs as a VM under vSphere, adding unnecessary compute means buying superfluous server resources -- and the software licenses to go with it.
Nutanix and a few other hyper-converged infrastructure vendors sell storage-only nodes for customers who want to add storage without the cost of vSphere licenses. But most HCI vendors don't support simply connecting external servers to the HCI appliance's storage to host more VMs. Customers must buy much more expensive HCI nodes, or in the case of VMware, buy a vSAN license for every server. NetApp executives say the company's HCI supports external servers.
Using VM hosts as storage nodes also creates operational complexity. In a traditional server and storage model, VM hosts are stateless. When a host needs hardware maintenance or a hypervisor upgrade, the admin simply migrates VMs off and upgrades the motherboard firmware. If that node is also used for storage, it has terabytes of state, and must be treated like a storage device.
VM hosts are clearly cattle, but even in a scale-out system, storage nodes are closer to pets. If a storage node goes offline, resiliency goes down, which increases the risk of data loss.
What Element OS has to offer
Another advantage of NetApp's approach to hyper-convergence is Element OS's quality of service (QoS). If NetApp migrated Element OS into a VM, that virtual machine would have to rely on the hypervisor for access to both the solid-state drives (SSDs) and the CPU. The hypervisor introduces latency to both SSD access and the processor, so Element OS couldn't provide the predictable latency that has been its biggest strength.
The downside to using Element OS and dedicated storage nodes is that the NetApp HCI minimum configuration is two appliances with four storage nodes: two compute nodes and two free slots. Because they share CPUs, traditional hyper-converged infrastructure systems can run effectively on two or three nodes. That makes them more cost-effective than NetApp HCI for small and medium-sized business and remote office/branch office applications.
NetApp HCI has three sizes of compute nodes, with 16, 24 or 36 cores and 256 GB, 512 GB or 768 GB of RAM, as well as storage nodes with six SSDs in 480 TB, 960 TB and 1.9 2 petabyte capacities. As with traditional SolidFire nodes, the storage nodes all include nonvolatile RAM (NVRAM). That should provide much better write latency than HCI products that use industry-standard hardware and that have to write data to the SSDs before acknowledging them.
NetApp built an automated installation application to make the setup process easier. That's a good fit for organizations that choose HCI to simplify their infrastructure and shorten the time between new kits arriving in the data center.
NetApp's HCI appliance isn't really hyper-converged, but that doesn't matter. The dedicated storage nodes deliver all the goodness of the Element OS, including QoS and the low write latency only NVRAM can deliver. The architecture also allows customers to buy servers and compute nodes at server markups and only pay the higher prices typical of storage systems and HCI nodes for their storage nodes. True HCI products remain attractive for remote offices, but the disaggregated model fits large data centers better.
Editor's note: According to NetApp, the company's HCI appliance will become available in Q4 2017.
Convergence improves storage function, affordability
Hyper-convergence brings cloud-style pricing to your data center