Microsoft Fabric: Infrastructure & Capacity
Infrastructure and Capacity in Microsoft Fabric
As a SaaS solution, Microsoft Fabric allows users to efficiently manage all relevant data workloads without the need for complex infrastructure provisioning. While Microsoft Fabric offers numerous advantages, proper capacity planning is particularly crucial for successful adoption. In this article, we take a detailed look at the Fabric infrastructure and show you how to avoid performance issues and unnecessary costs with the right capacity strategy.
How is the Microsoft Fabric infrastructure structured?
The Microsoft Fabric infrastructure is hierarchical and based on a clear structure of tenants, capacities, and workspaces, which together ensure efficient data management and processing.
Tenant: A tenant represents the top level of the hierarchy, typically corresponds to the entire company or organization, and is linked to your Microsoft Entra directory. Within a tenant, you can create multiple workspaces and capacities.
Capacity: A capacity is a dedicated pool of resources within your tenant. The size of the capacity, measured in Capacity Units (CU), determines the available computing power and directly influences the performance of your workloads. It plays a central role in resource allocation and pricing. A capacity can contain multiple workspaces.
Workspace: These serve as containers for various projects, teams, or departments and contain different items such as lakehouses, data pipelines, or reports. In this area, teams can collaborate on data, reports, and pipelines.
Capacity models in Microsoft Fabric
A central element of Microsoft Fabric is its capacity models, which govern resource allocation and usage.
F-SKUs
Capacity in Microsoft Fabric is defined by various F-SKUs, each providing a specific number of Capacity Units (CUs). These range from F2 (2 CUs) to F2048 (2048 CUs), allowing for flexible scaling based on your requirements. Since capacity is not pre-allocated to specific workloads, organizations can dynamically distribute their resources across different Fabric services. Two billing models are available, which can be purchased via the Azure portal:
- Pay-as-you-go: Billing based on actual consumption in CUs per second. Ideal for companies with variable usage.
- Reserved capacity: Pre-booking for 1 or 3 years with discounts of up to 41%. Suitable for companies with consistent usage and long-term planning.
Which capacity level is right for my company?
Choosing the right capacity depends on your company's specific requirements and workload. Actual capacity pricing also depends on the chosen data center.
For informed capacity planning, you should use the Microsoft Fabric Capacity Metrics app. This application helps you monitor resource usage and determine the optimal capacity size for your specific workloads.
As a rough guideline for capacity, you can consider the following:
Small Power BI projects and data engineering pipelines: For smaller BI projects, an F4 or F8 license may be sufficient. Depending on the use case, these may need to be scaled up at a later stage.
Real-time data processing: Applications with low latency and high concurrency, such as streaming data analytics or IoT data processing, benefit from an F256 or F512 capacity. If even faster processing is required, an F1024 SKU may be necessary.
Data science and machine learning: For AI models, neural networks, and deep learning calculations, an F256 is recommended. If compute-intensive training processes are performed with large datasets, an F1024 capacity may be required.
Since workload requirements vary, it is advisable to start with a smaller capacity and test scalability flexibly to achieve optimal performance.
How do I create a Microsoft Fabric capacity?
Scaling and capacity management
Microsoft Fabric uses automated mechanisms to process workloads efficiently. Bursting, smoothing, and throttling are three essential concepts that influence how compute capacities are utilized.
Bursting: Dynamic resource expansion
With bursting, the platform can utilize more resources than originally allocated during sudden load spikes to ensure consistent performance. This additional resource usage is treated as a "loan" from future capacity. This allows tasks that would normally take longer to be completed more quickly. However, it is important to note that these borrowed resources must be "repaid" later.
Smoothing: Even distribution of resource usage
To minimize the impact of bursting, Microsoft Fabric implements smoothing. This spreads the consumption of additional resources over a specific period:
- Interactive operations (Interactive Compute): Resource usage is smoothed over a minimum of 5 minutes.
- Background operations (Background Compute): Here, smoothing occurs over a 24-hour period.
This approach prevents short-term load spikes from overloading the capacity.
Throttling: Protection mechanism for overloads
Even with bursting and smoothing, it is possible for capacity to be overloaded over a longer period. In such cases, throttling acts as a protective mechanism:
- Interactive Delay:
If capacity is fully utilized for the next 10 minutes, new interactive requests will be delayed by 20 seconds. This applies, for example, to direct user requests such as Power BI queries or SQL queries in Lakehouses.
- Interactive Rejection:
In the event of sustained overload for 60 minutes, new interactive requests will be rejected. However, scheduled background operations, such as ETL pipelines or Spark jobs, will continue to run.
- Background Rejection:
If the overload persists for more than 24 hours, all requests—both interactive and background operations—will be rejected until resource usage falls back below the defined limit.
This tiered approach ensures that system stability is maintained and critical background processes can continue to run for as long as possible, while interactive requests are throttled early to protect the overall performance of the Fabric capacity.
Best practices for capacity management in Microsoft Fabric
Optimal capacity management in Microsoft Fabric requires strategic monitoring, performance optimization, and demand-based scaling. The following examples help to use resources efficiently and avoid bottlenecks.
1. Continuous monitoring and resource management
Structured monitoring of utilization allows for the early identification of performance issues.
- Fabric Capacity Metrics App: This app provides detailed insight into the usage of available resources and helps to identify capacity limits in good time.
- Capacity alerts: It is recommended to set up alerts that notify you of critical utilization spikes so you can react promptly.
- Workload management: Allocating capacity to specific workloads ensures that business-critical applications always have sufficient resources.
2. Optimizing workloads to boost performance
Efficient resource usage ensures improved query speeds and prevents unnecessary overloads. Here is how you can optimize your workloads:
- Optimized data models: Using compressed storage formats like Delta Tables and Parquet reduces resource consumption and speeds up queries.
- Partitioning large tables: Targeted partitioning of data volumes improves processing efficiency and reduces latency.
- Adjusting the storage strategy: Depending on the use case, a combination of DirectQuery for real-time analytics and Import mode for high query speeds may be appropriate.
3. Scaling capacity on demand
If performance bottlenecks persist despite optimization measures, a targeted expansion of capacity may be necessary. This includes, for example:
- Increasing capacity: If you consistently experience high CPU usage or memory bottlenecks, scaling up to a larger capacity tier may be required.
- Workload-specific adjustments: Limiting individual operations and avoiding resource-intensive processes during peak times helps distribute the load.
- Monitoring for scaling decisions: Regular analysis of capacity utilization helps determine the right time to scale.
These measures enable sustainable and cost-effective use of Microsoft Fabric, ensuring that resources are deployed efficiently and bottlenecks are avoided.
Success factors for optimal Microsoft Fabric utilization
Microsoft Fabric is a powerful, scalable data platform that combines flexibility with high performance. To unlock its full potential, thoughtful capacity planning, efficient workspace organization, and workload optimization are essential. Continuous monitoring helps you identify bottlenecks early and make necessary adjustments. It is best to start small and scale gradually as needed.
