Designing Microsoft Fabric for Scale

Session overview

Fabric makes it easy to create a workspace and start loading data. The decisions that matter later are made in those first weeks, often implicitly: how workspaces map to teams, where raw and curated data live, which engine writes which tables, and how changes reach production. This session makes those decisions explicit. It shows the trade-offs behind each one, so a platform can grow without being redesigned.

The session is architecture-first and vendor-neutral in tone. It covers what Fabric does well, where its boundaries are, and which choices are hard to reverse.

Outline

1. Boundaries before boxes: workloads, ownership, security and lifecycle as the basis for workspace design.
2. Lakehouse, Warehouse or both: write paths, transactions, team skills and serving needs.
3. Medallion without dogma: when three layers help and when fewer are enough.
4. Capacity: how compute is consumed, smoothing and throttling, and separating noisy workloads.
5. Large datasets: incremental processing, partitioning and file design from the start.
6. Change and operations: Git integration, environments, deployment and run logging.
7. A reference architecture: putting the decisions together, with the reasoning behind each arrow.

1. Sources

   • Databases
   • Files
   • APIs

2. Ingestion

   • Pipelines
   • Notebooks

3. Bronze

   • Raw Delta tables

4. Silver

   • Conformed entities

5. Gold

   • Serving models

6. Consumers

   • SQL endpoint
   • Semantic models
   • APIs

Cross-cutting concerns

• Git
• CI/CD
• Logging
• Monitoring
• Security

Adapting the session

• Talk or webinar: the architecture walkthrough, with diagrams and decision checklists.
• Workshop: attendees design a platform for a sample scenario in small groups, then compare their decisions with the reference architecture.
• Internal session: the same structure, applied to the organisation’s own workloads and constraints.