In the wake of the OpenStack Summit, held in Vancouver this year, two major questions remained. First and perhaps most obviously, why in the holy hell aren’t there more technology conferences held in Vancouver? Sure, it’s marginally more difficult to get into than San Francisco by air – at least if your primary carrier is JetBlue, which doesn’t service Vancouver. But this is the view from the conference center, which is itself quite impressive.
(click to embiggen)
Not that I have anything against California as a conference destination, mind. If Las Vegas is Mos Eisely, San Francisco is Shangri-La. But there is not a venue in San Francisco that can hold a candle to the Vancouver Conference Center and its absurd backdrop of mountains, water and lazily circling float planes.
Aside from the interesting but ultimately trivial question of venue, however, there was one big question following the Summit: what does the future hold for OpenStack?
The OpenStack project has always been a fascinating exercise in contradictions. On the one hand, it has attracted the kind of broad industry support and investment that other projects would kill for, and outlasted would be competitors like CloudStack and Eucalyptus. Both of which had distinct and theoretically marketable technical advantages over the offspring of NASA and Rackspace, notably. On the other hand, it is simultaneously and continually maligned by technologists from a variety of quarters. Much of this criticism is naturally the result of competitive messaging; in spite of its participation in the project, for example, VMware’s vCloud/vSphere teams predictably have less than kind things to say about OpenStack’s track record for success. But the criticism of OpenStack is hardly limited to competitors. Some of today’s largest contributors, in fact, initially passed on participating in the project after evaluating the first incarnation. And even significant OpenStack players today acknowledge that OpenStack has a lot of work to do moving forward.
None of which helps answer the question of where OpenStack is headed as a project, unfortunately. Technology is criticized all the time, frequently with merit, but the correlation of criticism with lack of adoption has not been strong, historically. Engineering quality matters, but not as much as the industry would like to believe. For better and, frequently, for worse.
The first question that needs to be unpacked when considering the fate of OpenStack is also one of the least interesting. Many critics of the OpenStack project build their arguments on fundamental questions of the economics of private versus public infrastructure. The arguments essentially boil down to an assertion that private infrastructure makes little sense for most companies – or any but the very largest internet companies, if you’re in the business of providing public infrastructure. These arguments tend to be built on macro-economic foundations: most private companies won’t be able to compete with the economies of scale realized by the public infrastructure providers, public cloud players will achieve outsized technical advantages through deeper vertical integration, and so on. I have made these arguments myself many times: see here for an example.
The simple fact is, however, that even if we assume, counter-examples like Etsy notwithstanding, these arguments to be entirely correct with the math unassailable and the downside risk of private investment perfectly understood, private infrastructure will be a fact of life for the foreseeable future. Whether or not it should be based on a rational, dispassionate evaluation of the variables is just not relevant.
Without descending into micro-economics in the form of the rational choice theory, the market evidence available to date is that in spite of subtantial and frankly unprecedented growth for cloud services, private infrastructure is a preferred strategy for many organizations that on paper would appear to be perfect fits for public alternatives. Whether these choices are rational or correct in an academic sense is, again, besides the point. The cloud is growing at an incredible rate, and the hits to x86 server manufacturers amongst others underscores that point, but there are still more businesses treating servers like pets than cattle. More importantly, there are legions of IT staffers that will be protecting what they believe is their livelihood – the private infrastructure – at all costs. Unless technical leadership is willing to wage total war on its own infrastructure, then, private infrastructure will continue to be a thing.
If we assume, therefore, that private infrastructure will remain a market – if one facing more competitive pressure than at any other time in its existence – the question is whether or not you want your private infrastructure to resemble public infrastructure in the feature sense. Do you want the provisioning of an instance to take a week or ninety seconds, in other words?
The answer to that question hopefully being self-evident, we’re left with two conclusions.
- First, that there will be private infrastructure on some reasonable scale.
- Second, that that infrastructure must be or become competitive with base level features of public clouds.
Which implies that there will be a market for private cloud software. This is the market that VMware has been steadily monetizing, and the market in which OpenStack is, with all due respect to Apache CloudStack, the last open source project left standing from a visibility standpoint.
All of which seems to be good news for the OpenStack project, and is, to some extent. There is competition on the way, certainly, from newer combinations such as Docker/Mesos and other private infrastructure fabric alternatives (in spite of complementary use cases today), but at least at present they require users to conceptualize their environment in ways the average OpenStack user may not be ready for. For today, anyway, the most likely t-shirt a customer wears to a meeting with an OpenStack vendor is one from VMworld. OpenStack has a window, in other words. How big that window is depends in part on one’s estimation of cloud growth and the downside impact to private competition, but it is difficult to build the case that the entire world will transition away from its private infrastructure to public alternatives in the short to medium term.
But keeping that window open depends on the answer to a fundamental question OpenStack has, and continues to, struggle with: what is OpenStack? The confusion around this subject manifests itself on several levels. Most obviously there is project composition: while OpenStack is typically referred to as if it were a single project, it is better described as a (growing) collection of independent sub-projects – some of which compete with one another. This in turn has several implications.
- The independent nature of the projects has made installation and upgrades problematic, historically.
- It is a burden for OpenStack vendors and marketers, who must educate would-be users of OpenStack on the nature of the project and the choices available to them.
- And finally it’s meant that defining what – precisely – a canonical OpenStack instance consists of has been a hard enough question that answering it has been a project of its own.
Those issues, however, are solvable problems. Or more correctly, they should be, except for the other major manifestation of definitional issues. OpenStack has assembled one of the most impressive rosters of member companies in the industry. The good news is that this has ensured a growing, vibrant community and excellent project visibility. The bad news is that the number of member companies guarantees that members will inevitably have different, and often competing, visions for the project’s future. It’s not difficult to understand that what a carrier might require from OpenStack, for example, could look very different from what an implementer would like to see. Neither of which is likely to be what an operating system vendor expects. And so on.
OpenStack is hardly the first open source project to be the center of broad-based, cross-category investment and collaboration, of course. Linux has been successful on this front for years. But successful projects have typically had a clear sense of purpose and identity: to be an operating system kernel, for example. OpenStack’s raison d’être, by comparison, has been less clear. On a high level, it’s been a mechanism for building an IaaS-type private cloud, but there have been major disagreements between project participants on how to get there, what to build it from and more. In some circumstances, this diversity can be a strength: the ability of OpenStack to substitute different storage subsystems based either on issues with the default choices, customer preferences or both has been useful. Abruptly attempting to redefine the IaaS vision to suddenly extend into PaaS, less so.
This existential crisis notwithstanding, if it is true that private infrastructure investments will be sustained over time, and that said infrastructure should borrow features from today’s public infrastructure, it is necessary to conclude that OpenStack has a market opportunity. Certainly the large system incumbents believe this. On the heels of 2014’s acquisitions of Cloudscaling (EMC), eNovance (Red Hat) and Metacloud (Cisco) – not to mention related pickups like Eucalyptus (HP), Inktank (Red Hat) and Nebula (Oracle) – consolidation continues in 2015. A few short weeks after the OpenStack Summit, Cisco and IBM announced separate OpenStack acquisitions within hours of each other in Piston and Blue Box respectively.
For these strategies to pay off, however, the OpenStack project and its members need an answer to the fundamental question of what is OpenStack. Without that, the project will have a difficult time improving the developer experience and will leave itself vulnerable to more focused projects with a clear sense of identity and purpose. Many in the industry laugh at the idea that Mesos, for example, is an OpenStack competitor, but the Mesophere tagline of “an operating system for your datacenter” would seem to put it squarely in contention for the private infrastructure opportunity OpenStack was built to address. OpenStack may today be more easy to adopt for enterprises given its resemblance to traditional infrastructure versus Mesos’ more forward vision of knitting many resources into a single large instance, but is that advantage sustainable?
This is but one of many questions OpenStack should be considering as it attempts to discover what, in fact, it is. The answers need to come soon, however, because the window will not remain open indefinitely.
Disclosure: Multiple vendors involved in the OpenStack project, including Cisco, HP, IBM, Oracle and Red Hat are RedMonk customers. VMware, which is both a participant in the OpenStack community and a competitor to it, is a customer. Mesosphere is not a customer, nor is the OpenStack Foundation.