Listen to this story
Kubernetes has never been better suited for cloud portability, and many IT buyers remain interested in the concept. But so far, there's little evidence that most mainstream enterprises are porting production apps between cloud providers -- and some companies are moving in the opposite direction.
Using Kubernetes platforms as a centralized management point for container clusters deployed among multiple cloud providers and edge locations has become standard fare for enterprise IT ops teams over the past five years. As early as 2019, at least a few early adopters also used Kubernetes clusters to facilitate moving production apps among private clouds, public cloud providers and cloud regions. These early adopters sought to keep workloads close to customers on global networks and ensure that they didn't get locked into one cloud provider.
Since then, more technical hurdles to cloud portability have been cleared, thanks to data storage products that improve the flexibility of stateful applications on Kubernetes, and fresh integrations between Kubernetes distributions, such as VMware Tanzu Application Platform's support for deploying apps on Red Hat OpenShift Kubernetes. The open source community has also created a production-ready API that standardizes how Kubernetes clusters are installed on a cloud provider's VM-based infrastructures.
Meanwhile, the results of a survey last year by Enterprise Strategy Group (ESG), a division of TechTarget, indicated that IT buyers' interest in cloud portability remained strong. Among 387 respondents to ESG's February 2022 cloud-native applications survey, 19% said that cloud-native app portability between data centers and edge locations, as well as between clouds, was critically important; 67% said it was very important, and 12% said it was somewhat important. Just 1% responded that it was not important.
However, that interest hasn't necessarily translated into real-world action yet. Analysts at firms such as Gartner and ESG said last month that they don't have research data available on how many enterprise IT orgs are moving apps from one cloud provider to another.
Despite advances in technology, many organizations still face cloud portability roadblocks, often because of a lack of IT skills in-house, said Paul Nashawaty, an analyst at ESG.
"The IT skills gap is a real issue -- only 5% to 10% of an organization, if they're lucky, have time to innovate, and they default back to the silos they know," Nashawaty said. "We predict that in the next 12 to 18 months, organizations will work with service delivery partners to achieve [cross-cloud portability] goals."
Kubernetes platforms cloud portability picture
Ultimately, there's much more involved in deploying apps on Kubernetes than just Kubernetes itself, and related layers of infrastructure, such as service mesh and event-driven architecture, have not seen the same industry standardization as container orchestration, said Gary Chen, a research director at IDC.
"At the beginning it was looking like, 'Oh, well, you know, there's gonna be a great world because everyone's gonna use Istio, and everyone's gonna use Knative' -- there were going to be more and more things that would automatically be part of the Kubernetes platform," Chen said. "But now it seems like the public cloud guys have all gone their own way on serverless … and it's not the same value proposition."
The recent rise of enterprise platform engineering has further complicated cross-cloud portability so far, Chen said.
Gary ChenResearch director, IDC
"The market is trending toward abstractions on top of Kubernetes with a better developer experience, rather than raw Kubernetes," he said. "That seems to be fracturing everything."
This fracturing is evident in recent survey data on Kubernetes platform adoption published in the Puppet by Perforce 2023 State of Platform Engineering report. Out of 438 respondents, 57% reported that their companies now have between two and four internal platforms, while 30% have five or more.
"[Portability] might be easier now at the cluster configuration level, but at that higher level, it doesn't seem like there's going to be a lot of consistency," Chen said. "One type of abstraction might win out [eventually] but right now, there are just so many different options out there, and there doesn't seem to be a clear leader."
Some enterprise IT teams with the skills to build their own Kubernetes platform out of raw, open source components might have laid the groundwork for cross-cloud portability. But for one such team at Liberty Mutual Insurance in Boston, there hasn't been demand for that feature, said Eric Drobisewski, senior enterprise architect for the company.
"We have not seen the need or the desire so far to have mobility across clouds," Drobisewski said. "In the industry, it's talked about a lot, but in practicality, it hasn't served a real purpose yet."
Instead, the Liberty Mutual platform team has begun to offload more of the management of Kubernetes cluster infrastructure to cloud provider services such as Amazon Elastic Kubernetes Service.
"We are trying to make that more of our common deployment model so that we can have a control plane that goes across multiple clouds at a visibility and a resource management level," he said. "We've set ourselves up well for [portability] when needed, but the focus has been on, 'Let's get the agility, the speed and some of the efficiencies we can [from managed Kubernetes] and then let's optimize that.'"
Cloud portability can thwart IT ops efficiency
Some Kubernetes platform teams have tried cloud portability, but have then gone on to make deeper commitments to a primary cloud provider instead.
One such IT organization worked for more than four years beginning in 2015 to deploy Kubernetes clusters across multiple clouds using Red Hat OpenShift Kubernetes as a common infrastructure layer. But about three years ago, the travel company in the Midwest drastically changed its plans, signing a 10-year agreement with Google that included a commitment to move all its apps to Google Cloud Platform (GCP).
"After a couple years, it became apparent that even having our infrastructure guys try to run [OpenShift] in three clouds was going to be a problem," said a senior director of enterprise architecture for that company, speaking on condition of anonymity because he did not have permission to represent it in the press. "Plus, on the business side, leadership was looking for somebody that we could have more of a partnership with, not just to handle the infrastructure layer, but to build products with."
Google's expertise in data science and AI were a selling point for the partnership and move to GCP, the senior director said. Mid-2022, the company shut down its last instances of OpenShift. It still has some apps in AWS but has moved most of its 50-plus applications into GCP, including into Google Kubernetes Engine.
"It became apparent that if you want to do multi-cloud, you have to boil yourself down to the lowest common denominator, bringing your own Kubernetes and a lot of your own stuff with you," the senior director said. "It was more important to us to get out of the physical data center world … and get quicker access to newer technologies in the cloud than worrying about being locked in with [a cloud vendor]."
Aerospace manufacturer Boeing is headed in a similar direction, though it will still run workloads in each of the major public cloud service providers, said Ricardo Torres, chief engineer of open source and cloud native and associate technical fellow at Boeing.
"Most teams shouldn't be doing multi-cloud," Torres said. "It's very cost inefficient and you should really understand why you need that before you go down that path … data egress [cost] is one [challenge]. … If you're trying to be multi-cloud, you really need to understand what the touch points are for your data."
As a company, Boeing doesn't necessarily plan to sign an exclusive contract with any public cloud infrastructure provider, but the platform teams Torres works with are looking for ways to consolidate IT automation tools and offload more infrastructure management tasks to a primary cloud provider's managed services.
This will help the company's platform engineers develop deeper expertise in how a particular cloud infrastructure works, rather than trying to maintain equal expertise across all major cloud providers, Torres said. That represents a shift away from an aversion to personal, informally maintained knowledge that's espoused by some Agile and DevOps practitioners, but this type of knowledge has its advantages, according to Torres.
"You get that shared experience of living that same thing day in and day out, rather than having everything shifted out from under you all the time," he said.
Beth Pariseau, senior news writer at TechTarget Editorial, is an award-winning veteran of IT journalism. She can be reached at [email protected] or on Twitter @PariseauTT.