Hyper-converged Is More Than an Architecture: It's a Lifestyle
Much more than a buzzword, it's the future of the datacenter.
Remember the days of sneaker-net and having to camp out at PCs to fix them? Not too long ago that was the norm, while today that's increasingly the exception. How about building computers by ordering a collection of parts? Or procuring a server just to run a single application? Before virtualization, that was the norm too.
Today in modern IT shops, a lengthy server procurement and deployment falls into the "I can't believe we ever used to do it that way!" category. With hyper-converged infrastructure and cloud computing solutions, we're reaching a similar inflection point. In short, hyper-converged solutions allow you to roll in anywhere, from rack mount servers to entire racks of servers that come with pre-baked software-defined data centers (programmatic compute, network, storage, and security). You simply plug in (power and network), boot up, do a simple configuration and you're all set.
Asking the Right Questions
Most IT leaders I meet with firmly hold the belief that highly agile and programmatic compute, network, storage and security will not create a competitive advantage. Instead, they simply will allow an organization to remain competitive over the next decade. That said, we should all be asking the question "Why am I building something custom that does not not create any competitive business advantage?" The answer to that question is usually "Because we've always done it that way."
It's hard to step outside a comfort zone, but why not spend less time on technology that's not differentiating and spend more time on technology that is? That's what innovation is all about, and all of us in the industry are innovators at heart. Further, there's a segment of the IT services industry that wants every IT organization in the world to have their own slightly customized infrastructure; doing so generates billions of dollars annually. In other words, there are plenty of people out there that have a financial interest in convincing you that hyper-converged solutions are bad for you.
Of course, select applications may require specialized hardware, but most applications are well-suited for a hyper-converged solution. Note that this isn't just a market centered on traditional datacenters; take a close look at the underlying architectures of every major public cloud provider and you'll see a highly orchestrated, robust, software-defined stack.
The Server Model
Rather than just talk about what hyper-converged is, let's talk about what we should expect from these solutions. In the long run, we should be able to think about purchasing hyper-converged systems the same way we look at buying servers today. You don't care how Intel and its partners architect the internal guts; the location of the processor on the motherboard is arbitrary. When buying a server, you worry about service-level agreements (SLAs), support and costs. Lock-in to Intel never enters your mind. That's due in part because Intel servers have standard interfaces like USB, Ethernet and PCIe. Switching from hardware providers is relatively painless outside of some potential changes in hardware management and associated processes. That said, let's look at what it would take to more comfortably feel the same about hyper-converged solutions as you feel about Intel-based servers.
For the sake of neutrality, Figure 1 shows a fairly generic hyper-converged system architecture. This represents an ideal hyper-converged model; each vendor in the market will have its own strengths and weaknesses. Programmatic compute, network, storage and security are fundamental necessities in any infrastructure; as a result, they should be what you demand from any hyper-converged offering. You could bring in multiple vendors to complete a stack, but that can create additional support challenges depending on how strong the hyper-converged vendor is at supporting other vendors' components.
Beyond how the system is plumbed, you should carefully evaluate the capabilities of the individual components, both in terms of features and automation capabilities. What storage capabilities exist, such as compression, deduplication, replication and quality of service (QoS)? What networking and security components are virtualized, distributed and automated (IP addresses, load balancers, firewalls), and can they traverse multiple datacenters and/or clouds? Ideally, the IaaS services offered by the hyper-converged platform should be universal to the point that applications can be managed across multiple datacenters and clouds, with as little friction as possible. I'm not saying that hyper-converged vendors should give you a single pane of glass for management, but their provided infrastructure services should offer some level of multi-cloud consistency.
Most importantly, carefully look at how the hyper-converged solution openly integrates with other services. For example, the solution may offer PaaS integration by supporting native Docker or Cloud Foundry APIs. Open IaaS integration could be provided by support for DefCore-certified OpenStack APIs. The solution should include extensibility to support third-party integrations as well. Open PaaS and IaaS support can give the platform the same amount of lock-in avoidance and integration flexibility as you get with better x86 servers to do a particular job today.
Let It Go
To be more modular and agile, you need to begin to let go of conventional infrastructure decision making and approach hyper-converged solutions with the expectations that you'll commit to them through a pre-determined refresh cycle. Modularity breeds simplicity and agility; once you start tinkering with everything in the module, you begin to trade off the agility and cost benefits, and potentially the reliability. Cloud providers don't offer the full keys to the infrastructure castle, because allowing customers unfettered access to everything related to infrastructure automation is a sure-fire way to introduce instability.
If you can get hyper-converged solution purchases to be as relatively simple as server purchases, you get a lot of time back to focus on business differentiation. This requires a commitment to treating infrastructure in a fundamentally different way, but will allow you to finally have time truly innovate at a pace your organization hasn't been able to realize in the past.
Learn more about "Software-Defined Shifts: The Business Value of Hyperconverged Infrastructure Solutions" on VMware Radius.
Chris Wolf is VMware's CTO, Global Field and Industry.