Can Direct Lake Semantic Model compete with SSAS Tabular Live Connection in Speed?

[Anusha66]

Can Direct Lake Semantic Model compete with SSAS Tabular Live Connection(which has pre-aggregated data available)? My concern here is the speed. 

Say we have one Fabric Semantic model(on F64 capacity) with 50 tables, and 200 measures and a Tabular SSAS cube with the same 50 tables and 200 measures and PBI Reports built on it.

Which report will have higher speed or how do we compare them.

[svenchio]

Hi @Anusha66  if you ever find who's the winner, please share it with me 🙏  ...  your question is VERY INTERESTING, I would be so happy to know the answer 😁 ! 

[jennratten · Accepted solution]

Hi  - thanks for posting to the community.  The short answer is, it depends  - maybe, maybe not.  

With the assumption that the direct lake model is well optimized, compressed, with v-order, etc.  (see links below)...

If both are called from a cold cache then SSMS will most likely be faster because direct lake needs to warm up.

Once the cache is warm they will likely be comparable; direct lake will perform very well, like import mode. 

There are a lot of other factor that play into this though.  The links below are good reads on points to think about and also demonstrate how to capture performance metrics for the semantic model.  Then you would also want to review the Fabric Capacity Metrics App to see how the tests impacted the capacity.

SSAS vs AAS vs Fabric Semantic Models: Ultimate Showdown!

Understand Direct Lake query performance - Microsoft Fabric | Microsoft Learn

Learn how to analyze query processing for Direct Lake semantic models - Microsoft Fabric | Microsoft...

(25) Choosing SSAS Over Power BI for Data Modelling (On-Prem) : Case Study | LinkedIn

Migrate from Azure Analysis Services to Power BI Premium: Migration scenarios - Power BI | Microsoft...

Microsoft Fabric: Not All Delta Tables Are Created Equally

Migrate Azure Analysis Services to Power BI - Microsoft Fabric | Microsoft Learn

@Anusha66

[Anusha66]

Thank you for providing the links!

[vivien57]

Hello @Anusha66 ,

These are two very different architectures in the sense that:

- with SSAS Tabular (on-premises technology), you have a network connection between the cloud and the on-premises world (or at least IaaS, which doesn't change much)

- with Fabric and a semantic model in F64, you will be in a cloud-to-cloud architecture and therefore have fewer network constraints.

In theory, the Fabric architecture is faster because there is no Cloud -> On-Premise network communication and, in addition, via OneLake, your data will be queried via DirectLake on PARQUET files.

In practice, you'll have to test it and let us know 🙂

Please feel free to give me a kudo if my answer helped you.

Have a nice day,

Vivien