OpenStack’s Dirty Little Secret: It Doesn’t Scale

OpenStack remains the open-source community’s cloud of choice, mustering tens of thousands of minions as it marches on toward cloudy relevance. That same community has turned OpenStack into a morass of competing projects and priorities, but there’s one area where OpenStack dearly needs even more community:

Helping it scale.

I talked with Mirantis co-founder and CMO Boris Renski this week in advance of his company’s announced partnership with Juniper Networks to improve OpenStack scalability. It quickly became clear that as popular as OpenStack is, it has a long ways to go before it’s truly enterprise class.

That is, without additional help.

Don’t Drink, Don’t Scale, What Do You Do? 

As popular as OpenStack has been to talk about, it has been less impressive in terms of deployments. As the October 2013 OpenStack user survey reveals, while companies were shifting from proof of concept to production, that production was relatively small.


Source: OpenStack User Survey, October 2013 

Similar data doesn’t appear to have been released in the November 2014 survey results, and perhaps with good reason: The scale isn’t very impressive.

In fact, as I’ve heard from a range of companies, a dirty secret of OpenStack is that it starts to fall over and can’t scale past 30 nodes if you are running plain vanilla main trunk OpenStack software.

That’s a pretty damning indictment, yes, though perhaps less cause for concern than originally appears. After all, as OpenStack pioneer Randy Bias has suggested recently, there’s no such thing as “vanilla OpenStack.” As such, he posits that those interested in running OpenStack should “Dial into the right level of ‘lock-in’ that you are comfortable with from a strategic point of view that meets the business requirements.”

Juniper To The Rescue

OpenStack adopters may cite “ability to innovate,” “open technology,” “cost savings,” and “avoiding vendor lock-in” as their top four reasons for embracing OpenStack, but the cloud technology’s inherent scalability problems means that anyone that wants to run it at significant scale is going to need to “dial into lock-in.”

See also: Software-Defined Networking: What It Is, How It Works, Why It Matters

As Renski tells me, one significant, but necessary, area of lock-in is giving OpenStack users a software-defined networking (SDN) fabric to deploy OpenStack clouds at scale. SDNs basically take over the role of directing network traffic between the physical servers used in cloud deployments, rerouting packets on the fly depending on demand and congestion.

Though Mirantis initially chose VMware to help power this with NSX, they’re now adding Juniper’s Contrail Networking, as well as OpenContrail for those that want to minimize lock-in.

While OpenContrail lags Contrail in terms of functionality, it’s still a step up from Mirantis’ past dealings with VMware. As Renski told me, “Companies needed NSX to scale. They had to pay the VMware scale tax. No longer.”

Of course, now they have to pay the Juniper scale tax, but at least they have choice, right? Renski continues:

NSX is an Achilles heel because Software-Defined Networking (SDN) and networking in the enterprise is typically a big decision traditionally driven by independent groups in the organization, not necessarily the same groups that are making OpenStack-related decisions. Many companies will choose NSX, but many will choose Juniper or other fabrics. And in cases where NSX was not chosen, VIO will end up getting locked out. Hence it is an opportunity for Mirantis to win more business as the Switzerland of Openstack by being pure play and partnering with Juniper Networks    

In other words, Mirantis is positioning itself to benefit whichever SDN a company may choose to use to scale OpenStack, though they don’t obviate the lock-in problem.

The Myth Of No Lock-in

Which isn’t really a big problem. No matter the software you choose, and whatever its license, there is always lock-in. As my former MongoDB colleague Vijay Vijayasankar notes:

Going back to Bias’ point, the trick is always to determine the amount of lock-in you’re willing to accept. For those companies that want OpenStack at scale, they’re going to need to embrace a certain amount of lock-in, at least where the SDN is concerned. 

Even if they elect to go with open source pure-play Red Hat, which I’ve encouraged in the past, the minute a company opts for a particular distribution, there is lock-in. It’s not a simple matter of rip-and-replace to dump one vendor for another, not in a world that has no vanilla distribution.

In short, you’re going to need to lock yourself into some technology to make OpenStack work for you. Better get used to it.

UPDATE: The original version of this article attributed the “dirty little secret” quote to Boris Renski, which is inaccurate. This was the author’s words, not Renski’s. We apologize for the confusion.

Photo by JD Hancock

Facebook Comments