7.3 Data Analytics Platform Design

Why Analytics Platform Design Closes Out Domain 2

The final data-storage skill, "Recommend a solution for data analysis," asks you to design the layer where raw and integrated data actually becomes insight — the data warehouse or lakehouse, the big-data processing engine, and the governance layer that keeps all of it discoverable, classified, and compliant. AZ-305 scenarios in this area typically describe a data volume, a workload shape (SQL-based reporting vs. data-science/ML), and a governance requirement, then ask you to choose the platform and defend the choice against cost or operational trade-offs.

Azure Synapse Analytics

Azure Synapse Analytics is Microsoft's unified analytics workspace combining SQL-based data warehousing, big-data (Apache Spark) processing, and pipelines in one place. Its two SQL query engines are a frequently tested distinction:

Synapse Spark pools provide on-demand Apache Spark clusters for large-scale data transformation, machine learning feature engineering, and unstructured/semi-structured data processing, auto-scaling and auto-pausing to control cost.

Table distribution strategy inside a Dedicated SQL Pool is also tested: Hash-distributed tables (data spread by a hash of a chosen column) are the answer for large fact tables in star-schema joins; Round-robin tables spread rows evenly with no logical grouping, good for staging tables before a full transformation; Replicated tables copy the full table to every compute node, appropriate for small dimension tables (under a few GB) that are joined frequently.

Azure Databricks

Azure Databricks is a first-party Azure service built on Apache Spark, offering a collaborative notebook environment tuned for data engineering, data science, and machine learning teams, with Delta Lake as its default storage format (adds ACID transactions, schema enforcement, and time travel to files in a data lake). Choose Databricks over Synapse Spark pools when the scenario emphasizes: a data-science-first collaborative workflow (shared notebooks, MLflow experiment tracking), advanced Spark tuning and Photon-accelerated performance, or an existing organizational investment in the Databricks platform and its Unity Catalog governance model. Choose Synapse (SQL pools + Spark) when the scenario emphasizes SQL-first enterprise reporting alongside big-data processing inside one consolidated workspace with tighter native integration to Power BI and Purview.

Governance Layer: Microsoft Purview

Microsoft Purview is Azure's unified data governance, cataloging, and compliance platform, and its role across the whole analytics estate is to answer "what data do we have, where does it live, how sensitive is it, and where did it come from" — regardless of whether the data sits in Synapse, Databricks, SQL Database, or on-premises. Tested capabilities:

• Unified Catalog / Data Map — automatically scans registered data sources (Synapse, Azure SQL, Cosmos DB, Data Lake, on-premises databases via a self-hosted integration runtime) and builds a searchable inventory of assets.
• Automated classification and sensitivity labeling — detects patterns like credit card numbers or national ID numbers and applies built-in or custom sensitivity labels, which can then drive Microsoft Purview Information Protection policies.
• Data lineage — visually traces how a dataset was produced and transformed across connected pipelines (including Data Factory and Synapse Pipelines runs), which is the specific capability to recommend whenever a scenario asks "how do we prove where this report's numbers came from" for an audit.

By 2026, Microsoft is directing new unified-analytics builds toward Microsoft Fabric, whose OneLake provides one logical data lake shared automatically across Synapse-style warehousing, Spark (via Fabric notebooks), Real-Time Intelligence, and Power BI, with Purview governance extending across it. For AZ-305, treat Synapse Analytics, Databricks, and Purview as the GA services the blueprint bullet still tests directly, while recognizing that a scenario explicitly asking for "one platform, no separate infrastructure to manage, native Power BI" is pointing at Fabric as the intended synthesis answer.

Exam Scenario Walkthrough

A retail bank needs: (1) a consistent, high-concurrency enterprise data warehouse for regulatory reporting queries running 24/7, (2) an ad hoc analytics zone where analysts occasionally query raw files sitting in a data lake without provisioning infrastructure, (3) a data-science team already standardized on collaborative Spark notebooks with MLflow, and (4) a way to prove data lineage for auditors. Recommended design: a Dedicated SQL Pool for the always-on warehouse (hash-distributed fact tables, replicated dimension tables), a Serverless SQL Pool for the ad hoc, pay-per-query analyst zone, Azure Databricks for the data-science team (keeping their existing MLflow/notebook workflow rather than forcing a migration to Synapse Spark), and Microsoft Purview connected across all three to provide the catalog, classification, and lineage the auditors require.

Takeaways

• Dedicated SQL Pool = pre-provisioned MPP warehouse billed by DWU for predictable heavy workloads; Serverless SQL Pool = pay-per-terabyte-scanned for ad hoc lake queries.
• Table distribution in Dedicated SQL Pool: hash-distribute large fact tables, round-robin staging tables, replicate small dimension tables.
• Choose Azure Databricks over Synapse Spark pools when the scenario emphasizes a data-science-first collaborative workflow, Delta Lake, or an existing Databricks/Unity Catalog investment.
• Microsoft Purview is the design answer whenever a scenario needs a unified catalog, automated sensitivity classification, or cross-pipeline data lineage for audit.
• Recognize Microsoft Fabric/OneLake as Microsoft's 2026 unifying direction beyond classic Synapse, but keep Synapse Analytics, Databricks, and Purview as the GA services the current AZ-305 blueprint tests directly.