How to approach hyper-converged architecture vendors and lock-in

freshidea - Fotolia

This article is part of our Essential Guide: Data center convergence and the role of storage

How HCI vendors threaten innovation opportunities

Deploying HCI gives IT shops one throat to choke when something goes wrong. But when a new tech advancement comes about, admins must wait on HCI vendors to catch up.

Hyper-converged infrastructure helps simplify data centers, but there is often a cost: HCI vendor lock-in.

Using hyper-converged infrastructure (HCI) doesn't require you to go all-in with one vendor, but the implementation details of HCI products often involve lock-in to get the most simplification and the best value. Part of your HCI evaluation should include whether the HCI vendors and products on your shortlist can easily support innovative technologies.

Many leading HCI vendors want you to replace your hypervisor hosts and storage with their appliances. If you follow this approach, you will be able to take advantage of every innovation your chosen vendor releases. But you will also need to wait for them to support anything new. That can mean waiting months for your HCI vendor to support a new hypervisor release that might be critical to your business.

At the other end of the scale are software-only HCI vendors. Products from these companies want to be layered on top of your existing IT infrastructure. In theory, you can use any new technology that works with the hypervisor and operating systems the HCI software uses.

Implications of HCI vendor lock-in

One trend that has emerged in hyperscale computing is the use of graphics processing units (GPUs) and field-programmable gate arrays (FPGAs) to solve complex problems on x86 servers, but this approach is at odds with HCI; hyper-converged infrastructure is based on the premise that an x86 processor is the best way to do any computing task.

A few HCI vendors have capabilities around GPUs for graphics acceleration, and they seem to recognize that other computing architectures are also important. But most HCI vendors probably won't add FPGA accelerator support. (It seems like FPGAs are turning up first as accelerated network cards. As long as HCI nodes have free PCI Express slots, the hypervisor should be able to use these cards.)

Another storage network feature you may want is non-volatile memory express (NVMe) over fabric, which supports newer shared storage. This is another technology that goes against the idea of HCI. Hyper-converged infrastructure uses storage inside the node rather than across a fabric (network) to a shared storage array. NVMe over fabric could show up inside HCI platforms as a faster way to connect nodes, but it is not the only storage innovation HCI could hold you back from. The proliferation of storage-class memory (SCM) is on the horizon.

Normal dynamic RAM DIMMs retain data only while the power is on, but SCM DIMMs keep their data; it's persistent immediately because it is written to main memory.

There are already DIMMs from Diablo Technologies that use NAND flash, although they are much slower than conventional RAM DIMMs. Intel and Micron Technology are next in line to deliver DIMMs based on their 3D XPoint technology. These new persistent DIMMs will be at least as revolutionary to storage performance as flash storage has been. Some applications will need to be rewritten to benefit from all the storage performance that will be available. But there's no way to know if your HCI vendor will support SCM down the line.

Convergence and hyper-convergence are all about simplification of both purchase and operations. But the business benefits that stem from these simplifications may be lost if they limit your ability to adopt new technologies. HCI vendors take different approaches to supporting new technologies in their products, and your needs and wants will change over time.

Next Steps

Hyper-convergence vendors provide disaster recovery

Mind performance needs to support big workloads on hyper-converged storage

How to choose an all-flash array vendor

Dig Deeper on Hyper-Converged Infrastructure Implementation