The Triad of Transformation: DevOps, MLOps, and ORAN in Modern Telco Infrastructure

What is DevOps ?

DevOps is a set of practices and principles that aim to improve collaboration and communication between development (Dev) and operations (Ops) teams in the IT industry. The primary goal of DevOps is to shorten the systems development life cycle, while also ensuring high-quality software releases and reliable infrastructure changes. It involves the automation of processes, continuous integration and continuous delivery (CI/CD), Agile processes and a culture of shared responsibility between development and operations teams.

Analogy with Cake Baking Process:

Let's try to understand this buzz word "DevOps' with baking a cake with a group of friends. Each friend has a specific role in making the cake:

Baker (Developer): This friend is responsible for coming up with the recipe, choosing the ingredients, and mixing everything together to create the cake batter.

Oven Operator (Operations): This friend is in charge of preheating the oven, putting the cake batter inside, and making sure it bakes at the right temperature for the right amount of time.

Now, let's see how this process compares to DevOps:

Collaboration: Just like in DevOps, the baker (developer) and the oven operator (operations) need to work closely together. The baker tells the oven operator what temperature and time are needed for the cake to bake properly.

Automation: In the baking process, imagine if the oven had a smart feature where you could set the temperature and time in advance, and it would automatically start baking when the time comes. This is similar to how DevOps uses automation tools to handle tasks like testing, building, and deploying software without manual intervention.

Continuous Integration (CI): Let's say the baker is trying out different variations of the cake recipe. With CI, after each attempt, the baker could give a slice of the cake to the oven operator for a taste test. This helps catch any issues early, just like in DevOps where developers frequently integrate their code changes, and automated tests are run to catch any problems early in the development process.

Continuous Delivery (CD): Imagine the baker is making multiple cakes for different occasions. Once a cake is ready and tastes good (passed testing), it can be delivered to the party. Similarly, in DevOps, when a piece of software is ready and passes tests, it can be deployed to the production environment, making new features available to users.

Feedback and Monitoring: If the oven operator notices the cake isn't rising properly or is browning too quickly, they can tell the baker to adjust the recipe. This feedback loop helps improve the cake. Similarly, in DevOps, monitoring software in real-time helps identify issues, and feedback loops help developers improve the code and fix any problems quickly.

Version Control: Imagine if the baker wrote down each cake recipe and any changes they made. This way, if they accidentally added too much sugar, they could go back to a previous version of the recipe. In DevOps, version control systems work similarly, helping developers track changes and revert to previous versions if needed.

Now imagine the same process without DevOps so how it work ....

Lack of Collaboration: The baker (developer) and the oven operator (operations) work independently. The baker hands over the cake batter without much communication about baking specifics.

Manual Oven Operation: The oven operator manually sets the temperature and time for each cake, based on their own judgment. There's no standardized process, so the oven settings might vary each time.

Limited Testing: The baker finishes making the cake batter and hands it to the oven operator, who places it in the oven. There's no tasting or testing during the process, leading to uncertainty about the cake's quality.

Inconsistent Results: The cakes come out of the oven with varying degrees of success. Some are overcooked, some undercooked, and some turn out just right. There's no systematic approach to ensuring consistent quality.

No Monitoring or Feedback: Once the cakes are baked, there's no way to monitor their quality or gather feedback from the guests. Issues might arise at the party, leaving the baker and the oven operator unaware.

No Version Control: If the baker wants to recreate a cake from a previous occasion, they struggle to remember the exact recipe and ingredients used, leading to inconsistency in taste and appearance.

In this scenario, the absence of a DevOps process leads to disjointed efforts, lack of collaboration, and inconsistent results.

What is MLOps ?

MLOps, short for "Machine Learning Operations," is a set of practices and principles that aim to streamline and automate the deployment, monitoring, and management of machine learning (ML) models in production environments. It bridges the gap between data science and IT operations, similar to how DevOps bridges the gap between development and operations in software development.

Key Differences netween DevOps & MLOps

While the foundation of MLOps is rooted in DevOps, there exist notable distinctions that render MLOps slightly different from DevOps, particularly in aspects such as data management and model training. The following diagram elucidates these key differences:

How DevOps and MLOps used in ORAN?

Now you might be thinking how this all process can work for RAN (Radio Access Network), well the answer is simple - thanks to the evolution of RAN towards more openness (OpenRAN) with the help of virtualized solution.

OpenRAN or ORAN is part of a standard under ORAN Alliance that defines an open and interoperable interface between different components of the RAN, such as the base station radio unit (RU), the distributed unit (DU), and the centralized unit (CU). This allows operators to mix and match hardware and software from different vendors and avoid vendor lock-in. vRAN is a virtualized RAN that runs the DU and CU functions on cloud servers instead of dedicated hardware. This allows operators to reduce costs, increase scalability, and optimise resource allocation. Cloud RAN is a cloud-native RAN that leverages containerization, microservices, orchestration, and automation to deploy and manage the RAN functions as software applications on cloud platforms. This allows operators to achieve faster innovation, better performance, and higher reliability.

DevOps and MLOps play an important role in enabling OpenRAN (ORAN). They help operators to automate the development, deployment, and operation of the RAN functions as software applications on cloud platforms. They also help operators to leverage ML/AI to optimize the network performance, enhance the user experience, and enable new services using service management and orchestration (SMO) functionality.

Service management and orchestration in O-RAN

The Open RAN Alliance (O-RAN) defines technical specifications and interfaces related to the RAN’s service management and orchestration (SMO) framework.

The SMO platform is an automation platform for Open RAN radio resources. Hierarchically, it is a component of the operational support system. Within the zero touch service management European Telecommunication Standards Institute (ETSI-ZSM) it is viewed as a RAN domain controller. The platform can be distributed, being deployed on premises, on the cloud or as-a-Service to suit the end-user requirements.

The ORAN defined SMO includes:

• A design environment for rapid application development
• A common data collection platform for management of RAN data as well as mediation for O1, O2 and A1 interfaces
• Support for licensing, access control and AI/ML lifecycle management, together with legacy north-bound interfaces (NBI)

SMO is an automation platform for Cloud RAN/Open RAN and purpose-built RAN. SMO offers platform capabilities using which automation applications can be built to support most common management use cases for ORAN networks. The application development and run-time environment in SMO offers a structured and standardized way to address these automation-led network operations:

• The design environment enables rapid application development and automation handled by DevOps
• The run time environment enables multi-vendor applications to run on the platform by offering capabilities such as Application life-cycle management , AI/ML training (using an MLOps process) and execution, Security, Data management, Policy, Analytics, Service and resource Orchestration etc. The capabilities are exposed to the ORAN applications (rApps) over the R1 interface.

The fusion of DevOps and MLOps principles within ORAN SMO and RIC (RAN Intelligent Controller) demonstrates a powerful approach for telecom. RIC acts as a platform for deploying AI/ML applications (xApps or rApps) on RAN functions, performing tasks like load balancing and anomaly detection. To succeed in this endeavour, operators follow these principles:

1. Version Control: Employ Git or SVN for meticulous tracking of code, data, and model changes in xApps or rApps.
2. Testing: Use tools like PyTest or JUnit to ensure functionality, accuracy, and performance of xApps or rApps.
3. Deployment: Containerization via Docker or Kubernetes streamlines packaging and distribution of xApps or rApps on cloud platforms.
4. Monitoring: Continuous oversight via Prometheus or Grafana monitors behavior, metrics, and errors, ensuring optimal performance.
5. Feedback Loop: Utilize A/B testing and online learning to gather user and network feedback, informing iterative improvements.

In essence, this integration optimises the development and operation of AI-driven applications in telecommunications, delivering robust and adaptable solutions that enhance RAN networks and communication services. It serves as a blueprint for orchestrating complex systems effectively.