第12章クラウドネイティブアプリケーション開発

Deployment technologies are broadly used in the industry today to deploy cloud-native applications. Containers/CaaS, PaaS, and FaaS/serverless deployment technologies are used by Netflix, Uber, and WeChat to deploy cloud-native systems that consist of many independent services. Knowing these systems, can you elaborate on the suitability of Containers/CaaS as compared to PaaS or FaaS/serverless for these particular systems?

There are various types of situations that prompt the use of cloud deployment technologies. The corresponding use cases and their implementation are covered in the following sections.

The most common types of deployments are for enterprise data centers, which will be the focus of these use cases. Typically, multiple cloud deployment technologies are used together to provide the services and deployment requirements to satisfy customer needs. As customers seek to deploy large-scale enterprise applications, they must assess the required infrastructure, middleware, runtime dependencies, and constraints to choose the most appropriate cloud deployment technologies. For cloud-native applications, cloud deployment technologies such as CaaS, PaaS, and FaaS are the most popular.

With bare metal servers, there is no virtualization and no hypervisor overhead, thus their resources (CPU, memory, and I/O) are dedicated to the application deployed on it. These resources, along with the hardware, are highly customizable and can be tuned up for maximum performance, making bare metal servers a viable option to support resource-intensive workloads such HPC. This also helps with latency-sensitive applications (e.g., streaming services) and large databases that consume significant resources. A combination of cloud deployment technologies can be used to satisfy the various requirements of a solution architecture. Bare metal servers can be used for the resource-intensive components, whereas VMs, containers, and other cloud deployment technologies deliver application services. Bare metal servers are also an option for legacy applications that may not be virtualized or not well suited for a virtualized environment.

Bare metal servers are also a good fit for security-oriented applications because additional security measures can be easily integrated. Bare metal servers are the best fit for solutions that must comply with regulations or contract agreements that require single-tenant hosting.

Legacy applications that can be virtualized but have known and manageable behavioral issues are well suited for cloud deployment using VMs. This also applies to legacy applications that are not well documented but there is basic knowledge of how they run. VMs are a viable option for monolithic and tiered applications that are not mature enough to be migrated to a microservices architecture. Such applications could be migrated to the cloud as cloud-based monoliths.

VMs can also be used to provide enhanced container security, although with additional overhead. The use of a VM can provide an additional layer of isolation when deploying containers eliminating the use of a shared kernel model.

Containers enable scaling, replicating, and choreographing services in a microservices architecture. Containers are also lightweight, portable, and platform independent. The ability to start containers in a fraction of the time of VMs is critical. With these advantages, cloud customers have developed a plethora of microservices ranging from user interface (UI) front ends to back-end microservices. A microservices architecture is a good use case for using containers.

Legacy applications that depend on out-of-support operating systems or other out-of-support dependencies are candidates for containers. Application dependencies can be containerized to enable execution. However, there are limitations to this approach, and it can be very challenging to get such applications up and running in a container. The application dependencies must be assessed to ensure containerization is a viable option. Legacy applications that can be modernized and migrated to the cloud are also good candidates for containers.

Platform flexibility is the primary use case for CaaS. Containers can consistently run anywhere without the limitations of middleware and platform tools offered by PaaS platforms. CaaS provides the customers with a container-based environment they can control (security, scalability, and availability) without significant effort.

The need to optimize VMs is the most common use case for unikernels at this time. Unikernels have smaller footprints, in part, because they are built without unnecessary binaries and libraries. Because they have less overhead, unikernels can start up very quickly, faster than VMs and containers. With the smaller footprint and speed comes improve security when it comes to unikernels. Because unikernels are built to include only the libraries needed to run the application, this reduces the cybersecurity attack surface, thereby making them more secure compared to the other cloud deployment technologies.

Mutable server infrastructures are where servers are continually updated and modified in place even after they are deployed. Immutable infrastructures are where servers are never modified after they are deployed. The main difference between mutable and immutable infrastructures is the policy used that determines whether components of the environment are designed to be changed after deployment. Customers may need or expect an immutable infrastructure to preserve the integrity of an environment. Although VMs are not permitted in an immutable environment, unikernels, by design, protect against such modifications because any updates or changes to any components are built from an image and provisioned to replace the old ones. Unikernels, thus, are a good choice for the need of an immutable infrastructure.

PaaS is used to deploy modern, cloud-native application architectures. PaaS provides a complete cloud platform that includes hardware, infrastructure, middleware, and runtime environment. It is most often delivered using a container-based infrastructure, and therefore supports the same use cases as containers.

The most prevalent use case for PaaS is speed to market. PaaS reduces or eliminates the burden of deploying and managing infrastructure, enabling cloud customers to focus instead on customer needs and development of the applications. PaaS can be more affordable, especially for software development houses, which often do not have the budget to hire personnel to build and manage infrastructure, as it offers access to a wider range of services up and down the application stack. PaaS also provides a cost-effective way for customers to scale up and down resources as needed. PaaS also provides a development and CI/CD environment to quickly develop, integrate, build, and deploy applications.

The use case for FaaS/serverless is limited to microservices-based applications, more specifically applications that are short-lived and event-based. Examples of services include data/stream and image processing, reservation handling, and IoT data processing. Other use cases include high-volume and parallel workloads and data aggregation tasks.

FaaS/serverless is used to reduce the customer’s responsibility to manage middleware dependencies, runtime considerations, and the underlying infrastructure. The FaaS/serverless platform takes care of this, leaving the customer with the responsibility to write the function code. The result is increased velocity in delivering highly available, technology capability, and satisfying business demand.