Choose a Storage Mode

This post is a part of the DP-600: Implementing Analytics Solutions Using Microsoft Fabric Exam Prep Hub; and this topic falls under these sections: 
Implement and manage semantic models (25-30%)
--> Design and build semantic models
--> Choose a storage mode

What Is Storage Mode?

In Microsoft Fabric, storage mode determines how a semantic model accesses and processes data. It affects performance, freshness, compute behavior, and model capabilities. Choosing the right storage mode is critical when designing semantic models for analytics and reporting.

A semantic model (Power BI dataset) can use different storage modes for its tables — and when multiple modes coexist, the model is called a composite model. DEV Community


Common Storage Modes

There are three primary storage modes you should know for the exam:

1. Import Mode

  • Stores data inside the semantic model in memory (VertiPaq) after a refresh. DEV Community
  • Offers fast query performance since data is cached locally.
  • Requires scheduled or manual refresh to update data from the source.
  • Supports the full range of modeling features (e.g., calculated tables, complex DAX).

When to use Import Mode:

  • Data fits in memory and doesn’t need real-time freshness.
  • You need complex calculations or modeling features requiring data in memory.
  • You want high performance for interactive analytics.

Pros:

  • Very fast interactive queries
  • Full DAX and modeling capabilities

Cons:

  • Must schedule refreshes
  • Data freshness depends on refresh cadence

2. DirectQuery Mode

  • Semantic model does not store data locally — queries are sent to the underlying source (SQL, warehouse, etc.) at query time. DEV Community
  • Ensures real-time or near-real-time data because no import refresh is needed.

When to use DirectQuery:

  • Source data changes frequently and must always show the latest results.
  • Data volumes are too large to import fully.

Pros:

  • Real-time access to source data
  • No refresh cycles required

Cons:

  • Performance depends heavily on source system
  • Some modeling features may be limited compared with Import

3. Direct Lake Mode

A newer, Fabric-specific storage mode designed to combine performance and freshness:

  • Reads Delta tables directly from OneLake and loads necessary column data into memory. Microsoft Learn
  • Avoids full data copy, eliminating the long import refresh cycle.
  • Uses the VertiPaq engine for fast aggregations and interactions (similar to import).
  • Offers low-latency synch with source changes without heavy refresh workloads.
  • Supports real-time insights while minimizing data movement. Microsoft Learn

When to use Direct Lake:

  • Working with extremely large datasets that would be costly or impractical to import entirely.
  • Needing relatively fresh data without long refresh cycles typical of Import mode.
  • Integrating tightly with delta-based assets such as Fabric lakehouses and warehouses. Microsoft Learn

Pros:

  • Fast querying with fresher data than import
  • No heavy refresh cycles
  • Leverages OneLake integration and existing delta tables

Cons:

  • Some modeling features (like calculated tables) are limited or not supported in Direct Lake tables (those tables must be switched to Import if needed). Microsoft Fabric Community
  • May fall back to DirectQuery in certain conditions (e.g., tables requiring SQL endpoint security). Microsoft Learn

Composite Models

A semantic model may include a mix of storage modes — for example, some tables in Direct Lake and others in Import. This is called a composite model. DEV Community

Typical use cases for composite models:

  • Import frequently used dimension tables (to support calculated tables)
  • Use Direct Lake for large fact tables stored in OneLake
  • Balance performance with modeling flexibility

Choosing the Right Storage Mode — Key Factors

When deciding on a storage mode for your semantic model, consider:

1. Data Freshness Requirements

  • Real-time data? → DirectQuery or Direct Lake
  • Static or periodic data? → Import

2. Dataset Size

  • Large volumes (multi-TB) without capacity for full import? → Direct Lake
  • Manageable size within memory? → Import

3. Modeling Features Needed

  • Complex measures, calculated tables, custom hierarchies? → Import (or mix)

4. Performance Needs

  • High interactive performance with good freshness? → Direct Lake
  • Ultimate speed with full caching? → Import

5. Source Capabilities

  • Some sources may not support DirectQuery efficiently — understand source performance.

Practical Examples

  • Import Mode: Small/medium enterprise data warehouse reporting that runs daily refreshes.
  • DirectQuery: Regulatory reporting where every query must reflect the latest operational data in a SQL system.
  • Direct Lake: Analytics on massive delta datasets stored in OneLake, where import is impractical but freshness and performance are both essential. Microsoft Learn

Exam Tips

  • Know what each mode does (Import vs DirectQuery vs Direct Lake).
  • Understand trade offs between performance, freshness, and modeling capability.
  • Recognize Direct Lake as a Fabric-optimized hybrid mode ideal for delta lake data.
  • Be prepared to choose the mode based on scenario requirements like latency, size, and features.

Summary

Storage ModeData LocationRefreshPerformanceBest Use Case
ImportIn model memoryScheduledVery fastSmaller datasets needing complex logic
DirectQuerySourceReal-timeSource-dependentReal-time needs
Direct LakeOneLake delta filesNear real-timeFast, scalableLarge datasets in OneLake Microsoft Learn

Practice Questions:

Here are 10 questions to test and help solidify your learning and knowledge. As you review these and other questions in your preparation, make sure to …

  • Identifying and understand why an option is correct (or incorrect) — not just which one
  • Look for and understand the usage scenario of keywords in exam questions to guide you
  • Expect scenario-based questions rather than direct definitions

1. Which storage mode stores data fully in memory within the semantic model?

A. DirectQuery
B. Direct Lake
C. Import
D. Composite

Correct Answer: C. Import

Explanation:
Import mode loads data into the VertiPaq in-memory engine inside the semantic model, providing the fastest query performance but requiring refreshes.


2. Which storage mode provides real-time access to data by querying the source system at query time?

A. Import
B. DirectQuery
C. Direct Lake
D. Cached

Correct Answer: B. DirectQuery

Explanation:
DirectQuery does not store data locally. Each query is sent directly to the source system, ensuring real-time or near-real-time results.


3. What is a key advantage of Direct Lake compared to Import mode?

A. Supports more DAX functions
B. Requires no OneLake integration
C. Avoids full data refresh while maintaining high performance
D. Works only with SQL Server

Correct Answer: C. Avoids full data refresh while maintaining high performance

Explanation:
Direct Lake reads Delta tables directly from OneLake, avoiding large import refreshes while still using the VertiPaq engine for fast analytics.


4. Which scenario is best suited for Import mode?

A. A dataset requiring real-time updates every second
B. A small to medium dataset with complex DAX calculations
C. A multi-terabyte lakehouse fact table
D. Streaming event data

Correct Answer: B. A small to medium dataset with complex DAX calculations

Explanation:
Import mode supports the full range of modeling features and offers excellent performance for datasets that fit comfortably in memory.


5. Which storage mode is specifically optimized for Delta tables stored in OneLake?

A. Import
B. DirectQuery
C. Direct Lake
D. Hybrid

Correct Answer: C. Direct Lake

Explanation:
Direct Lake is a Fabric-optimized storage mode designed to work directly with Delta tables in OneLake.


6. A semantic model includes some tables in Import mode and others in Direct Lake mode. What is this called?

A. Hybrid model
B. Incremental model
C. Composite model
D. Federated model

Correct Answer: C. Composite model

Explanation:
A composite model uses multiple storage modes within the same semantic model, allowing flexibility between performance and freshness.


7. Which limitation applies to Direct Lake tables?

A. They cannot be refreshed
B. They do not support relationships
C. Calculated tables are not supported directly
D. They cannot be queried using DAX

Correct Answer: C. Calculated tables are not supported directly

Explanation:
Calculated tables require Import mode. Direct Lake tables must be switched to Import if calculated tables are needed.


8. What primarily determines query performance when using DirectQuery mode?

A. The VertiPaq engine
B. The refresh schedule
C. The source system’s performance
D. OneLake caching

Correct Answer: C. The source system’s performance

Explanation:
In DirectQuery mode, queries are executed against the source system, so performance depends on source optimization and capacity.


9. Which storage mode minimizes data duplication while still offering high query performance?

A. Import
B. DirectQuery
C. Direct Lake
D. Cached Import

Correct Answer: C. Direct Lake

Explanation:
Direct Lake avoids copying data into the model while still leveraging in-memory query acceleration, minimizing duplication and refresh overhead.


10. You need near real-time analytics on a very large dataset stored in OneLake without long refresh times. Which storage mode should you choose?

A. Import
B. DirectQuery
C. Direct Lake
D. Snapshot

Correct Answer: C. Direct Lake

Explanation:
Direct Lake is ideal for large OneLake datasets where full import refreshes are impractical but fast, fresh analytics are required.

2 thoughts on “Choose a Storage Mode”

Leave a comment