Amazon Web Services (AWS) provides you with the flexibility to choose different compute options to build and run modern applications that map to your business needs. We provide you with access to the right operational model for your compute choice.
Developers and data engineers prefer a level of autonomy when choosing which compute models match which workloads. When initially developing modern applications, development teams need to manage and operate their applications directly.
As more workloads are developed, you might decide to create central platform or operations teams. The function of these central teams varies - some provide standard architectures and templated patterns for development teams to use, and others operate and manage workloads on behalf of multiple development teams.
These central teams strive to create standards for controlling costs, achieving the right performance and security, simplifying operations, and providing common architecture patterns. Achieving the right balance between autonomy and standardization is a challenge that enterprises and other large organizations often deal with.
Developers and data engineers may have different compute requirements. For example, a developer might choose AWS Lambda because it is optimized for event-driven patterns and gives access to hundreds of managed integrations.
Alternatively, a data engineer might choose an open-source framework like KubeFlow or Ray on Amazon Elastic Kubernetes Service (Amazon EKS) because it simplifies deployment of machine-learning models but allows access to the right high-powered instances.
Each role tends to develop skills in technology stacks over time, and has preferences about the tools they use, both look at their compute choice on a workload-by-workload basis, which means operational roles need to support a variety of workloads since they often work across teams. These roles include platform engineers, cloud administrators, or site reliability engineers (SREs).
Those in operational roles are challenged to provide autonomy for developers and data engineers while making sure they can deploy, operate, and monitor all workloads consistently to meet security, performance, resiliency, and cost requirements. Over time, as you develop more modern applications, these roles need to standardize on the tools to automate the deployment and monitoring of their workloads.
The choice between serverless compute and Kubernetes as an operational model is often driven by the need to have the right balance between autonomy and standardization with the number of resources one dedicates to running and operating workloads. Many workloads can be built successfully using either of these options. But, for some workloads, there are inherent advantages of one over the other.
A good team can make the compute choice entirely transparent to developers. Poor choices can limit developers options, and lead to sub-standard outcomes. Operators are always affected by the selected operating model and it will determine needs such as common libraries and networking configuration - as well as how the organization will interact with underlying services it needs to configure.
The operational model will often be determined by your organizational structure and skill around automation and operations. These roles might be distributed across development teams, different parts of your organization, or be centralized. The structure and skills of these teams are often among the most important considerations when choosing between a serverless or Kubernetes operational model.
The serverless operational model prioritizes shifting much of the work involved in provisioning and managing compute resources to AWS. This model includes services such as AWS Lambda and the Amazon Elastic Container Service (Amazon ECS).
The second model, Kubernetes on AWS, meets an important need for organizations that prioritize the Cloud Native Computing Foundation (CNCF) ecosystem with a managed experience using Amazon EKS.
For developers and data engineers, we recommend that you evaluate the most appropriate compute option on a workload-by-workload basis within your operational strategy. Here are some of the key criteria to consider when determining your strategy.
Organizational structure and skill
Enterprises tend to organize their development teams around one of the following models:
- Distributed: Each team owns their development, deployment, and operational functions. This model provides a great level of autonomy and allow teams to innovate at their own pace.
- Centralized: A centralized team maintains standards, creates automations, and facilitates the sharing of knowledge and best practices.
These are separate ends of a spectrum, and organizations may start with a distributed model and later migrate towards a centralized model as the number of workloads increases and the need for consistency across cost, performance and security of workloads becomes an important factor. However, in both models, serverless can reduce the infrastructure management overhead.
Another model organizations adopt, especially in the Kubernetes space, are the formation of central teams that specialize in platform engineering. These skilled engineering teams have developer skills and build and maintain platforms made up of common automation, deployment, and observability tools. Amazon EKS is often the choice for these organizations.
Identifying the structure and skill of your development teams can be critical to choosing the appropriate platform.
Organizations standardize on automation technologies to take full benefit of the cloud. The strategy and tools used by infrastructure, platform, and DevOps teams often drives decisions.
For example, organizations have choices for tooling that automates the creation, configuration, and maintenance of infrastructure, resources, and workloads. These tools include infrastructure-as-code (IaC) tools such as Terraform, CDK, CloudFormation, or other community tools.
Organizations that use Kubernetes might use Kubernetes-based tools for automation using GitOps tools such as ArgoCD and Flux and Kubernetes API-based cloud provisioning tools like AWS Controllers for Kubernetes (ACK) or the CNCF Crossplane project.
These tooling choices extend to tools that extend to security, testing, networking, observability, performance, and more. For each of these categories, there is a stack of available tools.
These automation tools often have built in integrations and accelerators for compute choices. For example, AWS Serverless Application Model (SAM) is optimized for serverless developers to build and quickly deploy Lambda functions. Other tools, like ArgoCD, automate the deployment and configuration of Kubernetes workloads.
As customers increase workloads, it can be a burden to support a large number of tooling choices. We recommend standardizing a set that support the most workload patterns.
It is important to evaluate the most appropriate compute option on a workload-by-workload basis within the default strategy. You should always strive to achieve the desired performance, security, and cost benefit for each workload.
A good standardization strategy will allow for different use cases such as microservices, modernized monoliths, event driven architectures, tools built by operation teams, and data processing workloads, such as machine learning, batch processing, and stream processing.
These workloads have different architectural characteristics. The strategy adopted should allow flexibility to support all the stages of a developer or data scientist workflow.
- Application Developer: Needs to run multiple environments such as development, prototyping, test, staging, and production.
- Data engineer: These workflows involve streaming and acting on large data models, cleaning, training, running inference with models, and building applications and data pipelines that experiment with the data, including Jupyter Notebooks.
- AI/ML scientist: Generative Artificial Intelligence (AI) Large Language Models (LLM) may require specialized compute instances such as AWS Trainium or other GPU-based architecture.
Applications do not exist in isolation. They are supported by technologies such as databases, messaging, streaming, orchestration, and other services. An effective modern app development strategy requires integration with these services. Managed integrations simplify operational overhead as much as management of the underlying infrastructure.
AWS serverless compute options, such as AWS Lambda, are integrated into the AWS ecosystem. AWS Lambda can subscribe to events from more than 250 other services.
AWS-managed offerings for Kubernetes also provide integrations with many AWS managed offerings. For example, you can use AWS Controllers for Kubernetes to provision native AWS resources using a Kubernetes API. In addition, Kubernetes has a rich ecosystem, offering integration with numerous open-source projects. Amazon EKS works with many other AWS services to provide additional solutions for business challenges.
Many organizations need to create experiments to validate ideas. The ability to provide an environment where you can quickly write, deploy, and validate ideas is essential for a healthy environment.
This environment is often overlooked when developing a modern app development strategy, but the ability to innovate may depend on it. Enabling teams to use services that allow builders to rapidly build, test, and iterate help in discovering new business opportunities and receive feedback faster.
Serverless compute options like AWS Lambda and AWS App Runner are optimized to enable organizations write code quickly, deploy it, and change it. These capabilities provide useful options for doing fast prototyping work that doesn’t require making a lot of choices upfront. Developers can quickly turn an idea into a modern, working application. The AWS serverless compute option provides choices to prototype with minimal cost or operational overhead.
Other options can be used for prototyping: for ECS, you can create dedicated clusters and use Fargate or dedicated nodes for prototyping, for Kubernetes, platform teams can create dedicated clusters or have namespaces within a cluster dedicated to prototyping teams. Open source projects such as CNCF BuildPacks or Knative can be used to simplify the configuration experience.
AWS offers different container options, such as Amazon ECS, serverless containers with AWS Fargate, and AWS App Runner, and different Kubernetes options, such as Amazon EKS, ROSA, and self-managed Kubernetes on Amazon EC2.
The following comparison table can help you determine your approach based on your workload requirements. You might choose pieces of both approaches, or have different teams that use different approaches. It is not uncommon to see a very large organizations have departments with different strategies.
Now that you have determined which approach best fits your workload for your environment, we recommend that you review the following service-specific resources to help you begin implementing your approach. This include links to in-depth documentation, hands-on tutorials, and other key assets to help get you started.
AWS App Runner
Getting started with Amazon ECS
We provide an introduction to the tools available to access Amazon ECS and introductory step-by-step procedures to run containers.
Tutorials for Amazon ECS
What's new and what's next with Amazon ECS
Learn what’s new since the launch of Amazon ECS Anywhere, new features of AWS Fargate, and a look ahead at the exciting enhancements to Amazon ECS.
Amazon ECS deployment
This guide offers an overview of Amazon ECS deployment options on AWS and shows how it can be used to manage a simple containerized application.
Amazon ECS workshop
This workshop is designed to educate those that might not be familiar with AWS Fargate, Amazon ECS, and possibly even Docker container workflow.
Deploy Docker containers on Amazon ECS
Learn how to run a Docker-enabled sample application on an Amazon ECS cluster behind a load balancer, test the application, and delete your resources to avoid charges.
AWS App Runner
Getting started with AWS App Runner
AWS App Runner: From code to a scalable, secure web application in minutes
Explore how AWS App Runner was designed to make it easier for you to deploy web apps and APIs to the cloud, regardless of the language they are written in, even for teams that lack prior experience deploying and managing containers or infrastructure.
Deploy a web app using AWS App Runner
Learn how to deploy a containerized web app using AWS App Runner. Start with your source code or a container image. App Runner automatically builds and deploys the web application and load balances traffic with encryption.
Getting started with AWS Fargate
Understand the basics of AWS Fargate, a technology that you can use with Amazon ECS to run containers without having to manage servers or clusters of Amazon EC2 instances.
Getting started with the console using Linux containers on AWS Fargate
Get started with Amazon ECS on AWS Fargate by using the Fargate launch type for your tasks in the Regions where Amazon ECS supports AWS Fargate.
Creating a cluster with a Fargate Linux task using the AWS CLI
Learn how to set up a cluster, register a task definition, run a Linux task, and perform other common scenarios in Amazon ECS with the AWS CLI.
What is AWS Lambda?
Learn more about AWS Lambda, a compute service that lets you run code without provisioning or managing servers.
Guide to AWS Lambda Pricing
Explorea and understand AWS Lambda pricing. You are charged based on the number of requests for your functions and the duration it takes for your code to start.
Using AWS Lambda with other services
Explore common use cases, learn how invocation works and includes a table that covers the services that work with Lambda and how it can be invoked from that service.
Getting started with Amazon EKS
Learn more about Amazon EKS, a managed service that you can use to run Kubernetes on AWS without needing to install, operate, and maintain your own Kubernetes control plane or nodes.
Amazon EKS deployment
Explore Amazon EKS deployment options on AWS and learn how it can be used to manage a general containerized application.
Amazon EKS Quick Start Reference Deployment
Using a Quick Start reference deployment guide, we provide step-by-step instructions for deploying Amazon Elastic Kubernetes Service (Amazon EKS) clusters.