Mathias Rosenthal - Fotolia
Enterprises that watch from the sidelines have a new milestone to note, as Kubernetes and Docker reach production on the bleeding edge of IT.
A rare bird has been spotted, and more are likely to follow -- a Kubernetes and Docker environment that is federated across multiple service providers' clouds and automates the distribution of container workloads based on data gravity and processing requirements.
Such Docker portability between clouds has been the promise of the container technology and the orchestration environments that manage it, but until the arrival of Kubernetes 1.5.1 last month, this multicloud orchestration wasn't practical, said Michael Bishop, CTO at Alpha Vertex Inc., a financial technology startup in New York.
The federated cluster felt like islands in previous versions, Bishop said, but now, it's more fluid with global deployment that routes a user to the nearest network ingress point. "We wanted to use the federation for high availability, so having an ingress that could point to wherever the deployments were was an important factor," he said.
A note of caution: While cluster federation has matured in the most recent Kubernetes release, Alpha Vertex has activated the feature before it's fully generally available.
"Some APIs are in beta, and others are in alpha," read the Kubernetes blog at the launch of Kubernetes 1.5 last month. "Some features are missing; for instance, cross-cloud load balancing is not supported."
Nevertheless, the startup's environment is at least one real-world demonstration of Kubernetes multicloud orchestration and Docker portability that could be imminent in the mainstream.
The cluster federation in Kubernetes makes workload federation easier through tagging, Bishop said.
"Using tagging, you're able to target where you want your job to run ... to your wildest imagination," he said. "You're able to express yourself just in one manifest [for multiple clouds], and it ends up where it needs to go."
Kubernetes and Docker deployment details
The firm creates big data models and uses machine learning to track financial market trends, with the goal to build a global graph of financial entities and their relationships. This more accurately reveals what actually moves financial markets, as opposed to what Bishop called "the CNBC version" of events.
Michael BishopCTO, Alpha Vertex
Alpha Vertex's platform ingests and inspects about a million pieces of information per day on a mixed infrastructure of Google Cloud Platform and IBM SoftLayer tied together with Kubernetes and Docker.
"We have some bare-metal machines [in SoftLayer] with a massive amount of RAM in them that we'd have a hard time getting in a virtual setting," Bishop said. For example, a test of a financial systems model from the graph against 30,000 to 50,000 stocks can take up to 9.75 million CPU core minutes.
A small portion of these models can be processed in parallel with a GPU. So, wherever possible, the data is shipped to suitable bare-metal machines in SoftLayer. The rest is processed through the auto-scaling Kubernetes cluster on Google Cloud Platform.
Real-world Docker portability for Alpha Vertex probably won't involve the live machine migration that's been the vision for technologies such as VMware long-distance vMotion.
"We do launch 'wherever there's room' jobs, but haven't had the need to drain a SoftLayer node and have pods pushed over to Google," Bishop said. "Even though the networks are all interconnected, we'd want to be cognizant of pushing work away from the data."
Some workflows in Alpha Vertex's environment may begin in Google, hand off to IBM's Bluemix OpenWhisk serverless computing service and end their journey as a job in SoftLayer. But each stage is self-contained and isolated from the next -- when one step ends, it posts its status to a message queue, and the next step begins.
"The future is multicloud, and having this federation allows you to use the best of whichever cloud provider's resources you need, without having to worry about setting up a new VM or a completely isolated cluster," Bishop said. "Each of the cloud providers has something unique about them, and if there's a very easy and portable way to leverage the best of each, I think more people should do it."
This type of deployment will become viable for a wider audience in the enterprise as the technology matures, analysts said.
Containers themselves tend to complement existing infrastructure or platform-as-a-service approaches, but "the same cannot be said for container management and orchestration software, such as Kubernetes," said Jay Lyman, an analyst at 451 Research.
Instead, container orchestration options will increasingly disrupt existing cloud management models, Lyman said.
This guide helps clear the multicloud air
Multicloud environments have dynamic benefits
Plan and manage your multicloud computing carefully