Nebula, maker of a hardware-based OpenStack turnkey solution, closed its doors on April 1 -- no joke.
The company, which was founded by one of OpenStack's original progenitors, failed in big part because it delivered a product that was less OpenStack than a proprietary solution.
"When we started this journey four years ago," reads a notice on Nebula's home page, "we set out to usher in a new era of cloud computing by curating and productizing OpenStack for the enterprise. We are incredibly proud of the role we had in establishing Nebula as the leading enterprise cloud computing platform. At the same time, we are deeply disappointed that the market will likely take another several years to mature. As a venture-backed startup, we did not have the resources to wait."
The maturity of the OpenStack market wasn't the only problem for Nebula. There was also the nature of its product stemming from expectations for OpenStack.
The OpenStack creators -- and the OpenStack appliance people
Nebula was co-founded by Chris Kemp -- formerly of NASA's Ames Research Center, where he developed part of OpenStack's core components -- with venture capital from Kleiner Perkins. His company's approach to the OpenStack market wasn't only to sell service or an enterprise distribution as Mirantis or Red Hat do, but to sell a complete turnkey hardware product.
The hardware in question was a controller appliance used to drive a cluster of up to 20 nodes per controller and up to five controllers per cluster. Setup was near-automatic, and upgrades to OpenStack were rolled out to the controller by Nebula.
Nebula believed its solution wasn't only more convenient, but "a true enterprise product with support that provides the enterprise integration capabilities (networking, storage, identity), all preconfigured and hardened," as Nebula CEO Gordon Stitt put it in an email earlier this year.
He believed Nebula's solution was superior to building OpenStack in-house and by hand, or hiring a professional services team: "We provide all of the testing and upgrading on the six-month cycle, and since our OpenStack implementation is a product, all of our implementations are identical so this is a simple, automated upgrade."
Unfortunately, the market didn't favor Nebula's hardware-centric approach.
The crucial mistakes
What went wrong? Cost of the solution aside, Nebula's product ran contrary to the way OpenStack itself was conceived.
Those who make a deep investment in OpenStack, as Walmart did, learn quickly how tough it is to work with. Standing up a stack and getting it running is only part of the deal; the work also includes upgrading the stack across releases, and -- most important -- building your own enterprise-specific customizations into OpenStack, which is one of the big reasons to invest in an OpenStack setup in the first place.
Once all that's mastered, the payoffs of using OpenStack also become bigger, and the resulting stack can become the enterprise's own creation, rather than the vendor's. With a managed, appliance-based approach, it's harder for a company to reintegrate its own changes.
Another place where Nebula fell down: It aimed for a shrinking market. Stitt saw Nebula serving enterprise that wanted to build "a truly seamless hybrid cloud using the same technology on-premise due to security or regulatory requirements and data residency requirements. The residency requirements -- that data never leaves the enterprise -- may be regulatory or may be due to the cost and/or latency of transferring data in and out of a public cloud."
But costs and requirements have been receding with each passing year, if not month. To that end, Nebula was not the only turnkey OpenStack solution out there, and many options were delivered as managed solutions on cloud-hosted hardware (Ubuntu BootStack, Mirantis OpenStack Express).
Finally, OpenStack's deployment issues seemed best solved within OpenStack itself through software -- which was bound to happen sooner or later -- rather than through what amounted to an ongoing dependency on a vendor's proprietary deployment solution.
What the market didn't bear
The wording in Nebula's farewell note seems to assign part of the blame on the way OpenStack has taken years to achieve the uptake its proponents envisioned.
This is in part a valid conceit. Only after four years did enterprise interest in the technology seem to turn the corner -- even then, the majority of its enterprise use, as of May last year, was modest compared to the scale and scope of Linux itself. Most of the deployments at scale were confined to verticals like telecom, with enterprise deployments largely modest affairs. (The most recent edition of the survey no longer tracks the size of deployments.) Container technologies also seem more likely to hit the sweet spot of what enterprises actually need, given the way they've enjoyed the kind of rapid and broad uptake that OpenStack hasn't seen.
But none of that has stopped companies like Canonical, Mirantis, or Red Hat from reaping investments made into OpenStack. With Canonical and Red Hat, their OpenStack work has been part of a larger portfolio of enterprise open source technologies, including containers. With Mirantis, it's been about easy deployment and a professed devotion to the openness of the underlying system.
In the end, the failure of Nebula is less about OpenStack's larger issues and more about a company that tried to sell a proprietary hardware solution to deliver and maintain an open source software stack -- in short, a company that provided the wrong solution to the wrong problem.