When GE Transportation wanted to move beyond its Java-centric development stack and containerize applications as much as possible, it knew it wanted to use Kubernetes, but also knew it needed help with it. It’s one of the customers touting Mesosphere DC/OS as the ticket to its Kubernetes transformation.
GE Transportation Digital Solutions operates a cloud development environment for developers and engineers creating applications for clients in the railroad, intermodal transport, and mining industries. The software applications can be a mix of cloud native, Java-based on-prem applications, and legacy apps, some of which are 20 years old.
It wanted a “container-as-a-service”-centric platform that it could offer in a more self-service way to support industrial applications in a cloud-agnostic and portable manner. The companies the team codes for need to blend incoming data from a variety of sources, including IoT devices on locomotives. That data is then used to drive analytics, network optimization, and transport logistic applications
“We can build an application that lives in a public cloud setup, but if our customer wants to run it on-premise, we want to make sure they can do so in a lightweight, cost-effective fashion,” said Wesley Mukai, Chief Technology Officer of GE Transportation Digital Solutions, in a case study.
“We provide them with a reference design, and they can stand it up themselves. In the future, our goal would be to create a more hybrid setup, with portability across different clouds and seamless on-premise portability. That’s an area where Mesosphere can really help us.” He added that the company also has more analytics projects in mind.
Mesosphere is among a quickly growing number of vendors focused on helping customers adopt a hybrid cloud, which it defines as a mix of data center and cloud, and multi-cloud, which involves multiple public clouds.
It recently released version 1.12 of Mesosphere DC/OS with Mesosphere Kubernetes Engine (MKE), which enables you to deploy multiple Kubernetes clusters on the same physical IT infrastructure. In a nod to the trend toward using its data center technology to manage resources across different clouds, it’s also rebranded DC/OS as Distributed Cloud Operating System.
After announcing last year that it would support Kubernetes in addition to its own Marathon container orchestrator, in February, it added deeper Kubernetes integrations and enhancements focused on edge and multi-cloud support, as well as plans for managed Kubernetes.
DC/OS is based on Apache Mesos, a platform for running any kind of distributed system. DC/OS also supports more than 100 data services, including Kafka, Jenkins and Spark.
“I think the reason cloud providers are really taking off is not because they’re good at providing VMs and storage — that’s been around for a while — but these services are becoming consumable on demand from a cloud provider. What we do at Mesosphere is provide you with that same level of automation and accessibility to those services, except we’re basically not tied to a cloud provider,” said Mesosphere’s Vice President of Product, Ed Hsu. “Any of these services, you can take with you to poly-cloud. You can have it run concurrently across various cloud providers with a unified multicloud operating model.”
In a recent survey, Kubernetes was the most-commonly used workload on Mesosphere, with 84 percent employing container/microservices architecture. Hybrid cloud was the fastest-growing deployment scenario.
DC/OS 1.12 includes a universal installer for cloud, multiple region deployment and operations capabilities for Kubernetes, improved LDAP user directory integration, metrics via Prometheus and more.
The company asserts that customers can cut costs by 50 percent or more with DC/OS and Mesosphere Kubernetes Engine (MKE). It offers a single control plane for the cloud, data center and edge; what it calls high-density multi-Kubernetes and improved lifecycle automation for security and Day 2 operations.
The single control plane allows users to manage multiple Kubernetes clusters across multicloud environments, multiple AWS zones, a co-location data center plus AWS and other scenarios.
It offers improved provisioning of the cluster, upgrading the cluster, “which is one of the hardest things to do — going from one version to the next,” according to Chris Gaun, product and marketing manager for Mesopshere, and automation like you’d get from a public cloud provider, but it’s software you can run anywhere.
“If a node goes down, it gets back up and running without any intervention from anybody in operations,” he said. “Imagine your Kubernetes control plane goes down. In a normal situation, this would be code red. Everyone would have to be on site to get it back up and running, and no other work would be done until that was done. With Mesosphere Kubernetes Engine you don’t have to do anything. It will come back up and running within minutes. We tested this out at scale, and the entire control plane took a couple of minutes to get back up and running.
With traditional Kubernetes, it’s one Kubernetes node per operating system – one Kubernetes node per VM, if you’re in a virtualized environment; per instance, if you’re in a public cloud; or server, if you’re doing this in a bare-metal environment Gaun explained. Companies often multiple Kubernetes deployments run by different teams that don’t talk to each other. This can lead to sprawl.
GE Transportation’s Mukai offered this:
“If a customer operates 30 terminals, then it may have 30 instances of an on-premise application running, and that’s very inefficient. …Even on-premise we want the software to be more consolidated and run more like a private cloud. Doing so makes it much more scalable when, say, the customer adds a 31st terminal to its operations.”
With high-density multiple Kubernetes, you can run multiple Kubernetes nodes on a piece of hardware, like a server, or a single virtual machine. That means, you could get rid of the virtualization, if you want to run on a bare-metal environment, and still have the isolation to run multiple Kubernetes components on a single bare-metal server, Gaun said. That would mean you could use fewer operating systems, less virtualization, less management overhead — you don’t have to have a cluster manager for each one of these clusters.
“If you consolidate all that and really think about all of the savings, it likely would be much, much greater [that 50 percent],” he said.
The company also added support for a beta release of a Mesosphere Jupyter service, which makes it easier for data science teams to share documents and visualizations using Jupyter Notebooks
Feature Image: “Railway Station Gleise Freight Trains” by fotolehrling. Licensed under CC BY-SA 2.0.