As software lifecycles continue to accelerate and as DevOps practices become commonly accepted, infrastructure and application deployments have become too complex and cumbersome to manage. You need to automate and standardize!
Agile Stacks DevOps automation platform makes it easy to implement infrastructure as code approach. In simple terms, Infrastructure as Code takes proven software development techniques and extends them to your infrastructure, allowing application environments to be deployed from code, and applying software development best practices to your infrastructure - like version control, automated testing, and continuous monitoring.
A Platform Stack is a collection of infrastructure services defined and deployed as a unit. Stacks are typically managed by automation tools such as Hashicorp Terraform, AWS CloudFormation, and Helm. Agile Stacks SuperHub is like a code generator for CloudFormation and Terraform. DevOps automation scripts can be generated and stored as code in a Git repository, avoiding the need to create Terraform and CloudFormation templates manually. Each component is stored in a separate directory with a deployment manifest that configures individual environments such as Dev, Test, Prod. In addition, Agile Stacks provides a catalog of integrated stacks of software, such as Kubeflow stack, Elastic stack, Kubernetes stack. This brings agility to DevOps teams by enabling them with highly scalable DevOps automation, using their selection of 50-plus components.
Similar to how DevOps accelerates software delivery, MLOps is a new discipline that enables successful operationalization and lifecycle management of your ML projects. ML applications frequently requires multiple complementary but independent processing steps (transformation, training, inference, etc.) and they require efficient collaboration between multiple roles: data engineers, data scientists, software developers and operations. Agile Stacks ML Pipeline Templates can be used to automatically create end-to-end pipeline for development of new models or to retrain/validate existing models based on new data. It's easy to get started with pipeline templates - just provide the location of your data, and let the pipeline retrain your model. In addition, data scientists can update pre-existing models using popular frameworks such as Keras and TensorFlow. Docker containers help to simplify development, code reuse, and deployment of each step in the machine learning pipelines. Containers provide scalability, better isolated test environments, and help to eliminate deployment issues caused by missing dependencies when deploying to a production environment. The same code that is used to train the model on a single CPU can be easily executed on a large number of parallel CPUs and GPUs, allowing data scientists to increase model accuracy and to complete experiments much faster - thanks to massive processing power of GPUs. Agile Stacks offers several templates for DevOps pipelines that leverage Docker and Kubernetes for your applications and machine learning workflows. DevOps pipelines for applications and machine learning can be maintained as code, allowing DevOps teams to implement immutable infrastructure, declarative security, unified logging, and automated deployments.
With agility and operationalization functionality provided by MLOps automation, data scientists don't need to become DevOps experts and instead can focus on problem analysis, data collection, and model development. At the same time, machine learning pipelines empower IT operations teams to take charge of ML applications in production with confidence and without having to know all the details of how the models work.
