Forum Discussion
Fabric Capacity Estimation
- 1 year ago
Hello sun-sboyanapall
FabianSchut shared a great guidance.
let me put my 2 cents here as wellWorkload Estimation
Your workload involves:
• ETLs: 50 ETLs with varying frequencies:
• 50% (25 ETLs) run hourly.
• 30% (15 ETLs) run daily.
• 20% (10 ETLs) run weekly.
• CU Consumption per ETL: Each ETL consumes approximately 1000 CUs per run.
From the calculations:
• Hourly CU Consumption: CUs.
• Daily CU Consumption: CUs.
• Weekly CU Consumption: CUs. Averaged per day: CUs.
The total daily CU consumption:25000*24 +360000+ 240000=1,200,000 cu
Monthly consumption:
12000000 * 30 =360,000,000 cU/month
F64 Capacity
The F64 SKU provides:
• Capacity Units (CUs):64*60*60*24=5,529,600 cu
Monthly capacity
5529600* 30 =165,888,000 cu
Your estimated workload (36M CUs/month) is well within the F64 capacity (165.9M CUs/month). This leaves significant headroom for additional workloads or unexpected spikes.
User Load and BI Reports
With around 100 users accessing BI reports:
• The F64 SKU supports up to 50 concurrent DirectQuery connections and a maximum of 30 live connections per second, which should be sufficient for your user base.
• The maximum memory per query is 10 GB, and the total memory for semantic models is 25 GB, which aligns with typical BI workloads.
Cost Optimization: Reserved pricing for F64 is recommended if the capacity will be utilized consistently.
• Feature Requirements: F64 supports advanced features like AI Skills and Copilot, which may be beneficial depending on your use case.Thanks
Your potential F64 SKU has 64 (total number of CU) * 3600 (seconds in an hour) = 230.400 CU second available. If you are running all your 50 ETL's every hour (max consumption), you would consume 50 * 1.000 = 50.000 CU second every hour. That means that you have 230.400 - 50.000 = 180.400 CU second left every hour to run all other tasks in Fabric, such as hosting the Power BI report. It depends on the sizes of your Power BI reports, but averagely taken, that should be enough. I think the Fabric F64 SKU is indeed viable for your scenario. Of course this is based on assumptions, like the sample dataset is an average of all ETL tasks. However, you have enough slack to safely say that it will be viable.