none
Question about Analysis Services Architecture (coming from SAP world) RRS feed

  • Question

  • Hi,

    I am coming from the SAP-world and need to understand how I can build a similar architecture in Azure Analysis Services like with SAP BW.
    In SAP BW you can pull the data directly from the source systems into a cube. Well, practically you don't, and instead build some layered architecture for transformations etc. before finally loading it into the cube. However, it is all based on the same RDBMS.

    Now looking at Azure Analysis Services, I understand that relational data storage (for transformations, e.g using Azure SQL DW) and OLAP (Analysis Services) is separated. I use any RDBMS to prepare the data, and then move it on into the OLAP engine from Analysis Services (which is also an SQL Server under the hood as far as I understand).

    So, if I already have cleansed and transformed data, can I directly ETL to Analysis Services and keep the data directly in the cube? Maybe also with some kind of disk offloading mechansim for saving money.

    And what about HANA? Is there something similar to an ADSO? (flat columnar in-memory, in contrast to a multi-dimensional in-memory)



    • Edited by a2r4 Sunday, October 28, 2018 12:52 PM
    Sunday, October 28, 2018 12:48 PM

All replies

  • Hi,

    In SAP BW you can pull the data directly from the source systems into a cube. Well, practically you don't, and instead build some layered architecture for transformations etc. before finally loading it into the cube. However, it is all based on the same RDBMS.

    ----In microsoft world also, you can directly pull data from source systems and create a cube on top of it, but as per the best practices and proven architecture it is always recommended and suggested to first clean & transform the data and then build dimensions and facts to create a cube on top of it. Trust me these concepts are independent of any technologies or tools. 

    So, if I already have cleansed and transformed data, can I directly ETL to Analysis Services and keep the data directly in the cube? Maybe also with some kind of disk offloading mechanism for saving money.

    And what about HANA? Is there something similar to an ADSO? (flat columnar in-memory, in contrast to a multi-dimensional in-memory)

    ---- Yes you can but as i said earlier you should model your data into dimension and facts before building cubes from source systems as those are highly normalized structures, but still if you don't want to do that then you can go with SSAS tabular models as they have columnar in memory capabilities and you can directly connect to relational source systems to build you models/cubes.


    Cheers,

    Monday, October 29, 2018 1:23 AM