“In OpenStack, code wins. That’s how it works. We’re not taking the position of, ‘Hey, we’re doing Kubernetes, and everybody else who was doing OpenStack around Swarm, stop what you’re doing!'” Renski says.
This goes beyond Mirantis tweaking its own software, however. (It’s actually more than a tweak; more about that later.) Mirantis and Google aim to rewrite Fuel, OpenStack’s deployment and managment tool, to use Kubernetes.
Their work will affect other OpenStack projects as well — most notably Kolla, the project to containerize OpenStack services. But the companies might also have to add pieces to Nova (OpenStack’s compute project) and Neutron (networking), Renski says.
By throwing down a bet on Kubernetes, Mirantis is potentially saving itself some engineering work when it comes to container support. “You can’t punch out a good product that is based on everything out there. Ultimately you have to make bets,” Renski says.
Mirantis wouldn’t be the first company to call Kubernetes a de facto standard. Kubernetes underlies Red Hat’s platform-as-a-service (PaaS), OpenShift.
But other options exist, including Mesosphere’s Datacenter Operating System (DCOS) and Docker Engine 1.12, which includes orchestration. And Startups such as Rancher Labs and newcomer ContainerX are also offering platforms for container management and operations.
That Mirantis envisions Kubernetes running on bare metal is key, because it will allow containers and virtual machines to co-exist. Mirantis envisions a data center running Kubernetes everywhere, with the option of running virtual machines (i.e. OpenStack) or containers on top of it.
“We’re enabling the customers to now have a single platform that is based on APIs for both containers and VMs, which is this Nirvana state our customers have been asking for,” Renski says.
An all-container world would be nice, but Docker and its ilk are still relatively new. Established applications run as virtual machines, so it’s necessary for any data center framework to support both structures.
Mirantis Goes CI/CD
In terms of a commercial product, the new Kubernetes-based OpenStack will arrive in Mirantis OpenStack 10, due for release late in the first quarter of 2017.
That’s later than Mirantis’ usual schedule of doing releases every six months; note that Mirantis OpenStack 9.0 came out just this month. But the delay is understandable given the fundamental nature of Mirantis’ changes.
In fact, once Mirantis 10 comes out, the six-month release calendar will be dead, Renski says. Mirantis will take advantage of Kubernetes to apply continuous integration and continuous development (CI/CD) methodologies to its software releases.
“We’ve containerized OpenStack, with Kubernetes as an underlay, so we can really implement a true CI/CD environment,” he says. After Mirantis 10, “We’ll be incrementally shipping just updated, patched containers.”
Why Google & Intel Even Care
It’s easy to see what Mirantis gets out of all this, but what about Google? According to Renski, it’s all about getting Google Cloud Platform some mindshare when it comes to on-premises networks.
Yes, that’s on-premises — meaning, non-cloud networks, or private clouds that aren’t farmed out to a managed provider. Microsoft has a foot in the door in the form of Azure Stack, an upcoming appliance that will essentially give enterprises their own offshoot of the Microsoft Azure cloud.
“Conquering that on-prem component is core,” Renski says. “Microsoft is launching Azure Stack, and there are all kinds of rumors that AWS [Amazon Web Services] is going to launch an AWS pod.”
That on-premises environment is based on virtual machines — which, let’s not forget, were the original purview for OpenStack, before this container trend exploded. So, through the Kubernetes project (which Google started), Google hopes to get a bit of mindshare in on-premises, virtual machine-based networks.
Kubernetes would also help OpenStack run microservices, Renski says. Microservices tend to be thought of as container-based constructs.
As for Intel, the chipmaker is interested in seeing on-premises clouds thrive, because that means a larger pool of potential cloud customers. “Intel doesn’t want to ultimately be selling their chips to, like, three hyperscale vendors,” Renski says.