Hyper-convergence: The Choice Is Yours

Vendors need to go beyond pre-packaged offerings, though.

Do you work for an "enterprise"? What does enterprise even mean to you? How do you judge it? Is your metric for "enterpriseness" based on number of employees, yearly IT spend or some other factor? If you do work for an enterprise, are your IT needs the same as other enterprises? Are all non-enterprises alike?

Leading questions, sure, but increasingly critical when talking about storage. The not-remotely-subtle point I'm driving at is that each organization is unique.

This isn't a popular idea amongst vendors: if everyone is unique, then actual effort needs to be put into dealing with edge cases and hand-holding administrators through support issues. This means it generally just isn't optimally efficient from a revenue standpoint, because vendors like nice homogenous groups of trouble-free customers with identical needs. Especially in storage.

In the oldentimes, customers having diverse needs was considered OK. Everyone just massively overprovisioned storage and forklift upgraded every few years. There weren't really any alternatives, so complaining didn't get the customer anywhere.
Things have changed.

The Emergence of Hyper-convergence
"Software-Defined Storage" (SDS) is a marketing term that has been largely rendered meaningless except as a catch all for "doing storage differently than we did it circa 2005". It covers categories ranging from scale-out storage to hyper-convergence. The net result of all that marketing and all those startups is that choice now exists in how storage is designed and deployed, and customers once again have the upper hand.

Hyper-convergence is among the more popular of the new storage architectures, and rightly so. Hyper-convergence offers several advantages:

  • Hyper-convergence can dramatically cut costs for organizations over traditional three-tier architecture for the simple reason that there are fewer bits to buy.
  • The storage lives in the servers that run the workloads. This also makes building the networks easier.
  • Workloads executing on a hyper-converged cluster generally execute on the same node that their storage lives on, providing that workload with very similar latency to having installed it on a dedicated bare-metal server. While not important for all workloads, this low latency makes many mission-critical ones (such as databases) very happy.
  • Hyper-convergence builds redundancy into the basic design; data lives on more than one physical node. If you walk up to one of the hyper-converged nodes in a cluster and shoot it, the workloads on that node would restart from data copies elsewhere. The list goes on and on, but the short version is that hyper-convergence makes storage easier and more affordable for many organizations struggling with traditional storage.
Reinventing Storage -- Again
Normally this would be cause for cheering; more diverse customer needs can be efficiently met than a decade ago. Of course, there are also a lot more customers needing proper storage than a decade ago!  With the increase in customer numbers has come an increase in diversity. Once more, customers are pushing the boundaries of "standard" storage designs.

The theory behind the core design advantages of hyper-convergence are that if you need more resources for your cluster -- storage, compute, what-have-you -- then you simply add another node. In order to avoid forklift upgrades, clusters can evolve over time and old nodes can be carefully phased out. In this manner, hyper-convergence helps to free customers from the tyranny of massive overprovisioning followed by forklift upgrades.

Reality, however, conflicts with this utopian marketing message. Hyper-converged vendors want to sell pre-canned appliances. Those appliances have a pre-set ratio of storage to CPU to RAM, and most vendors don't allow for a whole lot of deviation.

That thing where organizations are unique and have diverse needs? Hyper-converged appliances only solve some of the problem. There's a whole lot of the world that can't square the circle and are left paying for resources they'll never use.

Vendors, Tear Down These Walls
This is what SDS was supposed to be about: choice; ease of use; the ability to make a mistake or two in your datacenter design and not have to tear the whole thing up to fix it.

The future of storage rests not in replacing one set of design constraints with another, but in the relentless removal of barriers to diversity and the increased availability of customer choice.

About the Author

Trevor Pott is a full-time nerd from Edmonton, Alberta, Canada. He splits his time between systems administration, technology writing, and consulting. As a consultant he helps Silicon Valley startups better understand systems administrators and how to sell to them.


Subscribe on YouTube