Validating Findings on Optimizing Incremental Refresh with Microsoft Fabric

[goldenarm253]

Hello Microsoft Fabric Community,

I’d like to share some findings from our work on optimizing incremental refresh workflows using Microsoft Fabric Pro License Workspaces (Gen1) and Fabric F4 Workspaces (Gen2). The results have been promising, but I’d value input from the community to validate and refine these observations. While we’ve rated all aspects of our findings a 3 for now, we’re seeking validation and insights from the community to refine and confirm these observations.

Key Findings

1. CU Usage Reduction: By transitioning from Gen1 to Gen2 for incremental refreshes, we observed a 40% CU reduction, making the process significantly more cost-effective.
2. Performance Improvements:
   • Current setup (Gen2 only): Average refresh time is 3 minutes.
   • Proposed setup (Gen1 → Gen2): Refresh time improves to 1:45 minutes, a 41.67% reduction.
3. Strategic Workflow:
   • Using Gen1 for initial configurations (Pro License - no CU impact).
   • Leveraging Gen2 for optimized refresh cycles and workloads.

Community Feedback Request

To validate our findings and refine our approach, we’d like your thoughts on:

1. Have you observed similar CU usage reductions with a Gen1 → Gen2 setup?
2. Are there other strategies to improve refresh times or resource allocation in Gen2?
3. Does this approach align with your experience, or do you see potential improvements?

Diagrams below:

I rated everything a 3 for now.  

T

[v-nuoc-msft · Accepted solution]

Hi @goldenarm253 

Thank you very much for sharing!

1. In Microsoft Fabric, it is indeed a common phenomenon that CU usage decreases after transitioning from Gen1 to Gen2.

According to the official documentation, Gen2 is generally more efficient and can better utilize resources. This means that under the same workload, Gen2 may consume fewer computing units (CUs).

Differences between Dataflow Gen1 and Dataflow Gen2 - Microsoft Fabric | Microsoft Learn

2. In Microsoft Fabric, in addition to the strategies you have already mentioned, there are some other things you can try to improve refresh time and resource allocation in Gen2:

Optimize the data model to make sure it is as concise and efficient as possible. Remove unnecessary columns and tables, and optimize indexes and relationships to reduce processing time.

Set up incremental refreshes or partition large datasets so that different parts can be processed in parallel when refreshing.

Incremental refresh in Dataflow Gen2 - Microsoft Fabric | Microsoft Learn

Load data to Lakehouse using partition - Microsoft Fabric | Microsoft Learn

Continuously monitor system performance and resource usage and make adjustments. For example, allocate more resources during peak periods and reduce resource usage during off-peak periods.

Use the Monitor hub - Microsoft Fabric | Microsoft Learn

3. I think your approach is very promising in optimizing incremental refresh workflows. Some potential improvements are as I mentioned above, data model optimization, refresh methods, and monitoring resource usage.

Regards,

Nono Chen

If this post helps, then please consider Accept it as the solution to help the other members find it more quickly.

[goldenarm253]

Thank you so much for your detailed response and insights!

1. I’m glad to hear that the CU usage reduction from Gen1 to Gen2 aligns with Microsoft Fabric’s documented efficiencies. This validates a key aspect of our findings, and I’ll explore the resources you shared to further refine our approach.

2. Your suggestions for data model optimization, incremental refresh with partitioning, and performance monitoring are incredibly helpful. I especially appreciate the emphasis on simplifying the data model and leveraging the Monitor hub to track and adjust resource allocation effectively during workloads. These are actionable strategies I can incorporate right away.

3. I’m encouraged by your validation of our workflow and look forward to experimenting with the additional strategies you mentioned.

Thank you again for taking the time to share your expertise and for pointing me to the relevant documentation. I’ll follow up with any updates as we implement these recommendations!