Physical design

The physical design relates to the actual input and output processes of the system. This is explained in terms of how data is input into a system, how it is verified/authenticated, how it is processed, and how it is displayed. In physical design, the following requirements about the system are decided.

1. Input requirement,
2. Output requirements,
3. Storage requirements,
4. Processing requirements,
5. System control and backup or recovery.^[8]

Put another way, the physical portion of system design can generally be broken down into three sub-tasks:

1. User Interface Design
2. Data Design
3. Process Design

Web System design

Online websites, such as Google, Twitter, Facebook, Amazon and Netflix are used by millions of users worldwide. A scalable, highly available system must be designed to accommodate an increasing number of users. Here are the things to consider in designing the system:

1. Functional and non functional requirements
2. Capacity estimation
3. Database to use, Relational or NoSQL
4. Vertical scaling, Horizontal scaling, Shard
5. Load Balancing
6. Primary-secondary Replication
7. Cache and CDN
8. Stateless and Stateful servers
9. Datacenter georouting
10. Message Queue, Publish-Subscribe Architecture
11. Performance Metrics Monitoring and Logging
12. Build, test, configure deploy automation
13. Finding single point of failure
14. API Rate Limiting
15. Service Level Agreement

Machine Learning Systems Design

Machine learning systems design focuses on building scalable, reliable, and efficient systems that integrate machine learning (ML) models to solve real-world problems. ML systems require careful consideration of data pipelines, model training, and deployment infrastructure. ML systems are often used in applications such as recommendation engines, fraud detection, and natural language processing.

Key components to consider when designing ML systems include:

1. Problem Definition: Clearly define the problem, data requirements, and evaluation metrics. Success criteria often involve accuracy, latency, and scalability.^[9]
2. Data Pipeline: Build automated pipelines to collect, clean, transform, and validate data.^[10]
3. Model Selection and Training: Choose appropriate algorithms (e.g., linear regression, decision trees, neural networks) and train models using frameworks like TensorFlow or PyTorch.
4. Deployment and Serving: Deploy trained models to production environments using scalable architectures such as containerized services (e.g., Docker and Kubernetes).^[11]
5. Monitoring and Maintenance: Continuously monitor model performance, retrain as necessary, and ensure data drift is addressed.^[12]

Designing an ML system involves balancing trade-offs between accuracy, latency, cost, and maintainability, while ensuring system scalability and reliability. The discipline overlaps with MLOps, a set of practices that unifies machine learning development and operations to ensure smooth deployment and lifecycle management of ML systems.