Microsoft Fabric Architecture: The Unified Analytics Platform

Microsoft Fabric is the most significant change to the Azure data ecosystem since Azure Synapse Analytics. It unifies data engineering, data warehousing, real-time analytics, data science, and Power BI into a single, integrated platform — built on a shared storage layer called OneLake.

For enterprises already invested in Azure, Fabric simplifies a data architecture that previously required stitching together 5-10 separate services. Here's what it is, how it works, and when to use it.

What Fabric Actually Is

Fabric is a unified analytics platform that brings together seven workloads under one roof:

The unifying concept is OneLake — a single, hierarchical data lake that all workloads read from and write to. No more copying data between services, managing separate storage accounts, or building integration pipelines between analytics tools.

Architecture Overview

┌──────────────────────────────────────────────────────┐
│                    Fabric Capacity                      │
│                                                        │
│  ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│  │  Data     │ │  Data    │ │ Real-Time│ │  Power   │ │
│  │ Factory   │ │Warehouse │ │ Intel.   │ │   BI     │ │
│  └─────┬────┘ └────┬─────┘ └────┬─────┘ └────┬─────┘ │
│        │           │            │             │        │
│  ┌─────┴───────────┴────────────┴─────────────┴─────┐ │
│  │                    OneLake                         │ │
│  │         (Delta/Parquet, unified storage)           │ │
│  └───────────────────────────────────────────────────┘ │
└──────────────────────────────────────────────────────┘

OneLake: The Foundation

OneLake is Fabric's storage layer — think of it as "OneDrive for data." Key properties:

• Single copy of data. All workloads operate on the same data in OneLake. No ETL between analytics tools.
• Delta Lake format. All data is stored in Delta Lake (Parquet + transaction log), enabling ACID transactions, time travel, and schema evolution.
• Hierarchical namespace. Organised as Workspaces → Lakehouses/Warehouses → Tables/Files.
• Shortcuts. Virtual links to external data (Azure Data Lake Storage, AWS S3, Google Cloud Storage) that appear as if the data is in OneLake without copying it.

Lakehouses vs Warehouses

Fabric offers two approaches to structured analytics:

Recommendation: Use Lakehouses for data engineering and data science workloads. Use Warehouses for SQL-first analytics and reporting. Both read from OneLake, so you can use both without duplicating data.

When to Use Fabric

Strong Fit

• New Azure data platform. If you're building a data platform from scratch on Azure, Fabric is the default choice.
• Power BI-centric analytics. Fabric and Power BI are deeply integrated. If Power BI is your BI standard, Fabric is the natural data backend.
• Consolidating Azure data services. If you're running separate Data Factory, Synapse, and Data Explorer instances, Fabric unifies them with a simpler operational model.
• Microsoft 365 enterprise. Fabric licensing can be bundled with existing Microsoft agreements.

Weak Fit

• Multi-cloud data platform. Fabric is Azure-native. If you need cloud-agnostic data infrastructure, look at Databricks, Snowflake, or a custom Spark/dbt stack.
• Heavy real-time / streaming only. While Real-Time Intelligence is capable, dedicated streaming platforms (Kafka + Flink) offer more flexibility for complex event processing.
• Small data volumes. If you're processing gigabytes, not terabytes, Fabric's capacity-based pricing may be more expensive than simpler alternatives.

Licensing and Cost

Fabric uses a capacity-based pricing model — you buy Fabric Capacity Units (CUs) that are shared across all workloads.

Cost optimisation:

• Use pause/resume for development capacities (don't pay when not in use)
• Monitor CU consumption per workload to identify inefficient queries or pipelines
• Use OneLake shortcuts to avoid data duplication (storage savings)
• Right-size capacity based on actual utilisation, not peak demand

Migration from Existing Azure Data Stack

From Azure Synapse Analytics

Fabric's Data Warehouse is the successor to Synapse SQL. Migration path:

1. Create a Fabric workspace and warehouse
2. Migrate schema definitions (tables, views, stored procedures)
3. Point Power BI reports to the new Fabric warehouse
4. Migrate data pipelines from Synapse to Fabric Data Factory
5. Validate and cut over

From Azure Data Factory

Fabric Data Factory supports the same pipeline concepts (activities, triggers, datasets). Most pipelines can be migrated with minimal changes. Fabric adds native integration with Lakehouses and OneLake that simplifies pipeline design.

From Azure Data Explorer

Fabric Real-Time Intelligence provides equivalent capabilities with deeper integration into the Fabric ecosystem. KQL queries work in both environments.

Governance with Microsoft Purview

Fabric integrates with Microsoft Purview for data governance:

• Data catalog: Automatic discovery and classification of data in OneLake
• Lineage: Track data flow from source through transformation to reporting
• Sensitivity labels: Apply and enforce sensitivity labels across Fabric items
• Access policies: Centralised access management across workspaces

Common Mistakes

1. Over-provisioning capacity. Start small and scale up based on actual usage. You can resize capacity without downtime.
2. Ignoring workspace design. Workspaces are the primary boundary for access control and organisation. Design them around team ownership, not technology type.
3. Replicating old patterns. Fabric enables a lakehouse architecture that eliminates many traditional ETL patterns. Don't just move your existing pipeline complexity into Fabric — simplify it.
4. Neglecting monitoring. Fabric's capacity is shared across workloads. A runaway Spark job can starve your Power BI reports. Monitor CU consumption proactively.

---

Microsoft Fabric represents a fundamental simplification of the Azure data platform. If you're evaluating Fabric for your organisation or planning a migration from existing Azure data services, let's talk.