What’s holding Kubernetes back? As CNCF surveys reveal year after year, the top challenges enterprises face when using containers are still complexity, security, and monitoring, with recent surveys citing a lack of development team training and cultural change. Given the dramatic journey from monolith to microservices that Kubernetes represents, these challenges aren’t all that surprising. But some predict they’ll only get bigger, with Gartner predicting that by 2025, more than 95% of new digital workloads will be deployed on cloud-native infrastructure.
But the solutions are already here. From new software development approaches like internal developer platforms to innovations like eBPF that extend the cloud-native capabilities of the Linux kernel, exciting advances in cloud infrastructure are just around the corner. These industry-changing design patterns, open-source tools, and architectures will solve the complexity and scale problems of Kubernetes and advance cloud infrastructure as we know it today.
Reducing Cloud Native Complexity
Kubernetes needs to improve usability to thrive in the mainstream market. “Kubernetes is a wonderful standard API for accessing infrastructure across all clouds, but there’s a lot of work to be done to make it an enterprise-grade platform,” says James Waters, Broadcom’s director of research and development.
In the open source world, this problem is being addressed with internal developer platforms to reduce friction, and public cloud services offer solutions to make it easier to manage container infrastructure. Still, Waters believes there is a need for an enterprise application platform for containers that lowers the barrier to entry.
“Developers want access to self-service APIs, but they’re not always available at the lowest level,” Waters said. “APIs are not VMs-as-a-service or containers-as-a-service.” “Developers need more than just an application runtime-as-a-service to be productive.” Not to mention the major cloud providers, including VMware, Rafay, Mirantis, CubeSphere, and D2IQ, are working to make enterprise container management more usable.
Many experts agree that overall product complexity needs to be drastically reduced. “The complexity of cloud-native open source technologies is too high for the average enterprise,” says Thomas Graf, vice president of cloud networking and security at Cisco. He adds that compliance and security are common barriers to adopting cloud-native technology patterns in many on-premises brownfield situations.
Improved visibility into cloud resource usage
Most enterprises already use multiple clouds simultaneously. Analysts say this will become more common, and as a result, more cross-cloud management will be required. “A cross-cloud integration framework unifies data and workloads and operates collaboratively across clouds,” says Sid Nag, a VP analyst at Gartner. “This enables connectivity, adaptive security, and central management across all clouds.”
One way to increase awareness of activity in the cloud is to have an agnostic logging mechanism. “We’re starting to see the same energy we saw in Kubernetes in OpenTelemetry,” says Ellen Chisa, a partner at BoldStart Ventures. According to CNCF data, in mid-2023, OpenTelemetry was the second-fastest-growing project hosted by CNCF.
OpenTelemetry is becoming more important for several reasons. First, enterprises now have a lot of logs, and the cost of data is increasing. “As technology teams face real budget pressures from the board and the CFO, the question of ‘How can we make logging more useful to the business?’ is becoming more common,” Chisa explains.
Second, OpenTelemetry can enhance production environments with more context. “Just as you want ease of deployment (code to cloud), you’re going to want real-world information about what’s happening in the cloud when you’re writing code (cloud to code),” Chisa added.
Improved platform abstraction and automation
Today’s public and private clouds make IT infrastructure easier to use than ever before. Developers have more control with self-service APIs and user-friendly internal platforms. However, platform engineering still requires significant effort and needs to change.
The industry needs to move away from the YAML weeds and improve abstraction. “The next generation of serverless is one where the infrastructure is completely invisible,” says Jonas Bonner, CTO of Lightbend. Instead, he envisions a future where the actual operation of internal developer platforms is outsourced to operations teams or site reliability engineering (SRE) teams. “We’re in a transitional phase where developers and operations are learning to let go,” Bonner adds.
“Building enterprise-grade platforms remains labor-intensive, requiring significant effort to ensure the systems are secure and scalable,” said Broadcom’s Waters. “The platform team will play a critical role in infrastructure innovation by making it easier for developers to use in a pre-secured and pre-optimized way.”
Vercel CEO Guillermo Rauch emphasized that the latest frameworks “can completely automate the infrastructure.” Rauch expects a rise in framework-defined infrastructure and increased investment in the global front-end space. He also said that cloud infrastructure will evolve from being a bespoke, specialized infrastructure provisioned (and often overprovisioned) for each application to one that benefits both developer productivity and business agility.
Ultimately, it’s clear that streamlined internal platforms are where cloud infrastructure needs to go. “We’re moving toward an era where developers no longer have to worry about application functionality and dependencies, just as they no longer have to think about individual servers, data centers or operating systems,” says Liam Randall, CEO of Cosmonic. “Just as they expect the public cloud to maintain their data centers, they want their platforms to maintain common application dependencies.”
According to Randall, WebAssembly will usher in the next step in software abstraction and a new era beyond containerization. “Componentized applications based on the WebAssembly component model are compatible with, but not dependent on, container ecosystem concepts like service meshes, Kubernetes, and even containers themselves,” Randall explained. Components also solve the cold start problem, are smaller and more secure than containers, and enable “composable infrastructure” across language and language framework boundaries, he explained.
Introducing Virtualization to a Kubernetes Cluster
Another area of advancement is virtualization inside Kubernetes. “The same paradigm that drove hardware virtualization in Linux servers is now being applied to Kubernetes,” says Lucas Gentel, CEO of Loft Labs. The first is to address the ever-increasing costs of cloud computing driven by AI and machine learning workloads. In such scenarios, “sharing and dynamic allocation of compute resources is more important than ever,” Gentel explains.
The second reason is to address cluster sprawl. Half of the Kubernetes users surveyed by CNCF in 2022 were running more than 10 clusters. However, the number of clusters in use varies. Mercedes-Benz, for example, runs on 900 clusters. “Many organizations end up managing hundreds of Kubernetes clusters because there is no safe and simple way to achieve multi-tenancy within the Kubernetes architecture,” explains Gentel.
According to Zentel, virtual clusters can reduce the number of physical clusters required while maintaining the security and isolation required for diverse workloads, significantly reducing resource overhead and easing operational burden.
AI and Data Layer Orchestration
With the rise of AI, cloud-based infrastructure is expected to grow and evolve to meet new use cases. Gartner’s Sid Nag emphasized that “the combination of generative AI and cloud will be the next inflection point that will change the cloud infrastructure landscape.”
“Integrating specialized processors like GPUs, TPUs, and DPUs into the infrastructure will be key,” Nag said, adding that the ability to do this across different cloud environments, depending on unique AI requirements like training, inference, and fine-tuning, must also be addressed.
Orchestrating AI workloads is where Kubernetes excels. “Kubernetes will continue to be the go-to orchestrator for generative AI infrastructure,” says Rajeev Thakkar, director of product marketing at Pure Storage. Thakkar sees Kubernetes as an efficient way for data science teams to access GPU compute. But the sheer volume of data required for these models makes continuous access to persistent storage critical to their success, he adds.
Of course, managing stateful deployments in Kubernetes has been a tricky problem to solve for years, but the technology is now mature enough to overcome it. “It’s finally time for data in Kubernetes to go mainstream,” says Liz Warner, CTO at Percona.
“There’s still this perception that ‘Kubernetes is designed for temporary use and should be avoided,’” Warner explains. “However, today’s operators can reliably run open source databases like MySQL, PostgreSQL, or MongoDB with Kubernetes,” he adds, adding that this can lead to cost savings, better multicloud and hybrid solutions, and synergies across development environments.
Kubernetes on-premises and at the edge
Kubernetes and cloud-native technologies are starting to find a new home far from the cloud. “There’s this little-known magic sauce to Kubernetes,” says Cisco’s Thomas Graf. “It looks and acts very modern, but it’s backwards compatible, like a CPU, for 40 or 50 years.” Because cloud-native technologies are language agnostic, they can handle legacy code, making Kubernetes a natural fit for mass adoption. “Most enterprises are investing in this because they’re going to standardize on it for the next 10 years,” Graf adds.
“Containers in data centers are a relatively new concept, and this is going to evolve,” Graf emphasized. If the industry is going to move in this direction, it will need a modern, universal security mechanism that can avoid duplication of effort across both cloud and traditional data centers. Graf sees eBPF as a core foundation for a common networking layer and platform-agnostic firewall. eBPF is a secure, dynamically programmable way to the Linux kernel, made more accessible by the open source Cilium project.
The same shift is driving a new infrastructure paradigm at the edge. “A lot of the innovation over the last few years has been about decentralization,” says Jonas Bonner of Lightbend, noting the trend toward smaller instances of Amazon Relational Database Service and more powerful infrastructure to support users where they are: at the edge.
“It’s very wasteful to constantly be sending data to the cloud and back again,” Bonner says. “You need a platform where the data and the compute are physically right next to the end user.” This, Bonner says, provides the “holy trinity” of high throughput, low latency and high resiliency. This local-first development treats the cloud as a luxury for data redundancy rather than relying entirely on the cloud. As a result, Bonner explains, “the cloud and the edge are really becoming one.”
Bonner says that to realize this future of distributed hybrid architectures, a data fabric is needed. At the same time, the isolation that is a key consideration when moving data to edge devices makes WebAssembly a useful building block to replace containers. Lightweight alternatives to pure Kubernetes, such as K3 or KubeEdge, that can run cloud-native functions anywhere, will also be key.
Realizing the Future of Cloud Infrastructure
Kubernetes, the flagship cloud-native infrastructure, is poised to be used in more enterprise environments in the coming years, as are a host of innovations that continue to push the boundaries of what the cloud-native ecosystem can deliver: persistent data, cluster virtualization, platform engineering, logging, monitoring, multicloud management tools, and more.
What’s interesting is that as local compute improves and data ingress and egress costs increase, there’s a clear shift toward local-first development and deploying cloud-native tools to the edge and traditional data centers. “This brings a whole new level of complexity that’s largely unknown in the cloud-native world,” Cisco’s Graf said.
Generative AI will also bring amazing capabilities to this space, automating more and more cloud engineering use cases. “I think Kubernetes in particular will be a system that is going to improve incrementally, but it will be greatly empowered by AI,” says Omer Hamerman, Zesty’s lead engineer. “AI will make a quantum leap in the automation of Kubernetes and container-based application deployments.”
Other technological innovations are poised to reinvent much of what we take for granted across software development. For example, Cosmonic’s Randall notes that edge vendors are using WebAssembly to achieve higher levels of abstraction on their developer platforms. “A WebAssembly-native orchestrator like WasmCloud can auto-scale common plugin functionality across a variety of physical architectures and existing Kubernetes-based platforms,” Randall says. “With WebAssembly components, the future is already here. It’s just not evenly distributed.”
This seems to be a good summary of the entire cloud infrastructure. The future is already here, and much of it is based on progressive open source technologies. Now it’s time to make it happen.
editor@itworld.co.kr
Source: www.itworld.co.kr
