Hyperconverged infrastructure will evolve into cloud building blocks
- Summary:
- Hyperconverged architecture is wasted in on-premise deployments. Here's why.
Hyperconverged infrastructure (HCI), those dense 1-2U devices melding server, disks and software defined storage and management software into an easily deployed appliance, are a hot technology industry segment. In an era when server revenues have limped along with low-single digit increases, sales volumes of hyperscale systems have been on a rocket ship trajectory. While IDC estimates that overall server revenue increased 5.2% in Q4 of 2015, sales of what it calls "density-optimized" servers, a reasonable approximation for the hyperscale market, ballooned over 30%, contributing two points, or 40% of the overall increase.
Anecdotal confirmation comes from two primary sources. At a recent EMC World press briefing, Chad Sakac, head of its Converged Platforms Division noted that EMC's converged infrastructure line is now a $5 billion business growing at 20% annually while its newer hyperscale products like VXRAIL and VSPEX have already hit a billion dollar run rate with revenue accelerating at 50-60% CAGR. Meanwhile, the pre-IPO darling of the hyperconverged business, Nutanix recently reported revenue growing over 80% annually.
These extensible appliances melding standard server and storage components with virtual software stacks have captivated IT buyers by being easy to deploy, scale and manage. Despite the superficial resemblance of products like HPE Hyper Converged System, Nutanix NX and VXRAIL to the type of high-density systems used at Google or Facebook, they’re typically used for mainstream IT tasks like virtual server consolidation, VDI or scale out shared storage, not cloud-native software stacks.
While understandable given the still nascent state of enterprise private cloud implementation, it’s both unfortunate and likely to change since HCI is the (near) perfect platform for software-defined cloud infrastructure.
VMware’s well-known dominance of the enterprise virtualization market means that almost all HCI products are built on, sometimes exclusively, a vSphere hypervisor substrate. This is understandable since according to a 2015 survey about IT hyperconverged use and plans, 93% of respondents currently run vSphere with Hyper-V a distant second.
The result is that HCI is most often deployed as a purpose-built virtualization appliance to run traditional workloads, replacing discrete servers and storage arrays. However the platform’s key attributes, namely pre-integrated simplicity, scale out expandability with centralized manageability, make HCI well suited for IaaS and PaaS software stacks. What’s lacking is a vendor willing to take the lead by building an HCI/cloud stack bundle.
Rack-scale systems point the way
Since cloud services need a certain scale before making technical or economic sense, most vendors target rack-scale converged systems as the preferred cloud-in-a-box hardware platform. For example, Oracle’s Private Cloud Appliance integrates 2 to 25 compute nodes, redundant cloud controllers, storage and a virtual network fabric in a pre-built rack.
Likewise, EMC's recently announced VxRack Neutrino targets cloud-native, service-based applications via a turnkey implementation of OpenStack with a host of software add-ons for infrastructure management, monitoring and inventory, workload automation and capacity planning wrapped in a Web UI. EMC’s strategy for its converged infrastructure (rack-scale) business rests on the assumption that IT organizations would rather consume services than build infrasditructure, yet that's the same motivation for hyperconverged buyers.
Although EMC's rack system sales growth indicates it’s a successful reading of the market, Sakac expresses some uncertainty in this post explaining Neutrino writing,
[It] Will be interesting to see customer reaction – will they evolve past 'play' into 'consume/buy' for Cloud Native apps like we see them doing for traditional enterprise apps?
I believe they will, however the number able to drop 7-figures on a brand new, rack-scale cloud installation is a small subset of the market. Instead, HCI, with its much lower entry price and pay-as-you-grow scalability would be a better cloud platform for most organizations to build upon. In differentiating HCI and rack-scale, Sakai says:
Both are delivered in a turn-key fashion, it’s more about 'start small and scale out' versus 'start by thinking about HOW to scale really big and then scale out from there.'
There are several problems with this logic:
- Most organizations won't need a multi-rack cloud implementation, particularly in an era of 4-node HCI appliances, each capable of hosting hundreds of virtual CPUs
- Even organizations that require massive cloud scale typically don't want to build it in elephant-sized chunks
- HCI products like Nutanix actually don't have hard limits on the number of nodes in a cluster, plus its management system can handle multiple clusters from a single console
Thus, it doesn't make technical or business sense to limit turnkey cloud products to mainframe-sized racks.
The idea of a cloud appliance isn't new, Nebula pioneered the conceptfive years ago, but ultimately failed. Yet this was more a reflection of being too early to market, in which neither IT, nor the underlying technology was quite ready for private clouds, rather than a faulty product architecture.
More recently, Mirantis has worked with several hardware vendors to produce turnkey OpenStack installations at sub-rack scale. HCI is the logical architectural extension in which each multi-node chassis includes compute, storage and cloud control resources that can be amalgamated into larger clusters as needed.
My Take
HCI will evolve from a platform for conventional VM workloads to being the building blocks for IaaS and PaaS stacks. As with HCI storage clusters, it's probably infeasible to build a usable cloud with a single appliance, even one with four nodes, however it should be possible to start much smaller than today's rack-scale products. Besides, if you only need a few nodes, it's far easier and cheaper to use a public cloud.
Although the insta-cloud market has been tested with OpenStack, that isn't the only, nor necessarily even the most interesting cloud-native software platform. Expect savvy HCI vendors to also build products tailored for Azure Stack, containers (Docker, LXC, Rocket), CloudFoundry (on a variety of hypervisors) and other PaaS. The scale-out architecture and inherent technical and business benefits of HCI are too good to waste solely on legacy workloads.