ETL involves extracting data, then transforming it, and finally loading it into the target system, whereas ELT involves extracting data first, then loading it into the target system, and finally transforming it.

A broadcast variable in Apache Spark is replicated to every executor node for efficient data distribution. It's used for broadcasting large read-only datasets to all tasks across the cluster to avoid excessive data shuffling.

Data modeling tools like ERWin or Visio primarily aid in visualizing and designing Entity-Relationship Diagrams (ERDs), which depict the entities, attributes, and relationships in a database schema.

Direct extraction involves querying a database directly to retrieve specific data sets based on defined criteria. This method is often used when only a subset of data is required for analysis or processing.

Leveraging physical data modeling involves optimizing table indexes, partitioning data appropriately, and organizing the physical layout of data to enhance performance and address specific performance issues in the database application.

In an RDBMS, a view is a virtual table that represents the result of a database query. It provides a way to present data in a structured manner without storing the actual data, thus simplifying data access and enhancing security.

A conceptual data model represents the high-level structure and relationships between data entities. It is independent of any specific database management system and focuses on the business concepts and rules.

Dimensions in Dimensional Modeling are categories or entities that provide context to the facts stored in the Fact Table. They contain descriptive attributes that help in analyzing and understanding the data.

The primary difference between batch processing and streaming processing in pipeline architectures lies in the timing of data processing. Batch processing involves processing data in discrete chunks or batches at scheduled intervals, while streaming processing involves continuously processing data in real-time as it becomes available. Batch processing is suited for scenarios where data can be collected over time before processing, whereas streaming processing is ideal for handling data that requires immediate analysis or actions as it arrives.

Features such as DAG scheduling, task dependencies, monitoring dashboard, and retry mechanism in Apache Airflow would be particularly useful in scheduling and monitoring daily ETL jobs. DAG scheduling allows defining workflows with dependencies, task dependencies ensure tasks execute in the desired order, the monitoring dashboard provides visibility into job status, and the retry mechanism helps handle failures automatically, ensuring data pipelines complete successfully.