Both OpenStack and Docker were conceived to make IT more agile. OpenStack has strived to do this by turning hitherto static IT resources into elastic infrastructure, whereas Docker has reached for this goal by harmonizing development, test, and production resources, as Red Hat’s Neil Levine suggests.
But while Docker adoption has soared, OpenStack is still largely stuck in neutral. OpenStack is kept relevant by so many wanting to believe its promise, but never hitting its stride due to a host of factors, including complexity.
And yet Docker could be just the thing to turn OpenStack’s popularity into productivity. Whether a Docker-plus-OpenStack pairing is right for your enterprise largely depends on the kind of capacity your enterprise hopes to deliver. If simply Docker, OpenStack is probably overkill.
An open source approach to delivering virtual machines
OpenStack is an operational model for delivering virtualized compute capacity.
Sure, some give it a more grandiose definition (“OpenStack is a set of software tools for building and managing cloud computing platforms for public and private clouds”), but if we ignore secondary services like Cinder, Heat, and Magnum, for example, OpenStack’s core value is to gather a pool of hypervisor-enabled computers and enable the delivery of virtual machines (VMs) on demand to users.
Not that this is a small thing. After all, without OpenStack, the hypervisor sits idle, lonesome on a single computer, with no way to expose that capacity programmatically (or otherwise) to users.
Before cloudy systems like OpenStack or Amazon’s EC2, users would typically file a help ticket with IT. An IT admin, in turn, would use a GUI or command line to create a VM, and then share the credentials with the user.
Systems like OpenStack significantly streamline this process, enabling IT to programmatically deliver capacity to users. That’s a big deal.
Docker peanut butter, meet OpenStack jelly
Docker, the darling of the containers world, is similar to the VM in the IaaS picture painted above.
A Docker host is really the unit of compute capacity that users need, and not the container itself. Docker addresses what you do with a host once you’ve got it, but it doesn’t really help you get the host in the first place.
A Docker machine provides a client-side tool that lets you request Docker hosts from an IaaS provider (like EC2 or OpenStack or vSphere), but it’s far from a complete solution. In part, this stems from the fact that Docker doesn’t have a tenancy model.
With a hypervisor, each VM is a tenant. But in Docker, the Docker host is a tenant. You typically don’t want multiple users sharing a Docker host because then they see each others’ containers. So typically an enterprise will layer a cloud system underneath Docker to add tenancy. This yields a stack that looks like: hardware > hypervisor > Docker host > container.
A common approach today would be to take OpenStack and use it as the enterprise platform to deliver capacity on demand to users. In other words, users rely on OpenStack to request a Docker host, and then they use Docker to run containers in their Docker host.
So far, so good.
If all you need is Docker…
Things get more complicated when we start parsing what capacity needs delivering.
When an enterprise wants to use Docker, they need to get Docker hosts from a data center. OpenStack can do that, and it can do it alongside delivering all sorts of other capacity to the various teams within the enterprise.
But if all an enterprise IT team needs is Docker containers delivered, then OpenStack — or a similar orchestration tool — may be overkill, as VMware executive Jared Rosoff told me.
For this sort of use case, we really need a new platform. This platform could take the form of a piece of software that an enterprise installs on all of its computers in the data center. It would expose an interface to developers that lets them programmatically create Docker hosts when they need them, and then use Docker to create containers in those hosts.
Google has a vision for something like this with its Google Container Engine. Amazon has something similar in its EC2 Container Service. These are both API’s that developers can use to provision some Docker-compatible capacity from their data center.
As for Docker, the company behind Docker, the technology, it seems to have punted on this problem. focusing instead on what happens on the host itself.
While we probably don’t need to build up a big OpenStack cloud simply to manage Docker instances, it’s worth asking what OpenStack should look like if what we wanted to deliver was only Docker hosts, and not VMs.
Again, we see Google and Amazon tackling the problem, but when will OpenStack, or one of its supporters, do the same? The obvious candidate would be VMware, given its longstanding dominance of tooling around virtualization. But the company that solves this problem first, and in a way that comforts traditional IT with familiar interfaces yet pulls them into a cloudy future, will win, and win big.