Container orchestration vs. VM orchestration in the cloud

Question

I'm trying to understand the specific use cases where we'd prefer to use container orchestration (Kubernetes) as opposed to VM orchestration (Nomad) in an unmanaged cloud setting (e.g., EC2)

It seems to me that clearly, if you're focused on batch jobs, you're working with single-purpose VMs that are started then destroyed after doing their specific bit of work, so setting up a VM image to provision them with everything they need would seem to me to introduce less overhead into the cluster, and it wouldn't make much sense to use Kubernetes for a case like this. The distinguishing properties of the cloud that makes it easy to find one or more VMs that match the required scaling seem to me to make it as elastic and malleable as a container-level orchestration.

Take for instance the main reason I think people would use Kubernetes: autoscaling. why does VM-level autoscaling, as provided by e.g. Nomad, not suffice? Specifically in the cloud setting, the ability to create (with Packer), scale (with Nomad), and delete VMs is as straightforward as the Kubernetes case. I totally understand that a Kubernetes solution heavily favours cases where we're running on bare metal, but in the cloud setting, there's not much difference between starting/running/deleting containers with Kubernetes vs. starting/running/deleting virtual machines with e.g. Nomad.

In what specific cases would you prefer to use Kubernetes in an unmanaged cloud setting, e.g., EC2?

Answer 1

While both Kubernetes and VM orchestration tools like Nomad can manage resources effectively in an unmanaged cloud setting such as EC2, there are specific use cases where Kubernetes might be preferred:

• Containerized Microservices: Kubernetes is designed to manage containerized applications at scale. If your architecture is based on microservices or applications running inside containers, Kubernetes provides robust features for deployment, scaling, and managing these services efficiently.

• Service Discovery and Load Balancing: Kubernetes has built-in features for service discovery and load balancing. It automatically assigns DNS names to containers and balances the traffic across them. This simplifies the setup and management of complex network configurations.

• Resource Utilization: Kubernetes offers advanced scheduling capabilities that optimize resource utilization. It can pack containers onto nodes more efficiently, leading to better resource utilization than traditional VM-based orchestration.

• Automated Rollouts and Rollbacks: Kubernetes supports automated rollouts and rollbacks of application updates. This feature ensures that updates are deployed gradually, minimizing downtime and enabling quick rollbacks in case of issues.

• Declarative Configuration: Kubernetes uses declarative configuration files (YAML or JSON) to define the desired state of applications and infrastructure. This approach simplifies configuration management and promotes consistency across environments.

• Ecosystem and Community Support: Kubernetes has a large and active community with extensive documentation, tutorials, and third-party tools. This ecosystem support can be valuable for troubleshooting, integration with other systems, and staying updated on best practices.

• Horizontal Autoscaling: While Nomad also supports autoscaling, Kubernetes provides more fine-grained control over scaling based on metrics such as CPU and memory usage. Kubernetes Horizontal Pod Autoscaler (HPA) automatically adjusts the number of running pods based on observed metrics, ensuring optimal resource utilization and performance.

• Stateful Workloads: While both Kubernetes and Nomad can manage stateful workloads, Kubernetes offers more advanced features for stateful applications, such as PersistentVolumeClaims and StatefulSets. These features simplify the management of databases, message queues, and other stateful applications in a containerized environment.

While Nomad can effectively manage VM-based workloads in the cloud, Kubernetes offers additional features and capabilities specifically tailored for containerized environments. So if you or your organization is working with containerized microservices architectures, dynamic scaling requirements, and a need for advanced orchestration features may prefer Kubernetes in an unmanaged cloud setting like EC2.